From 388b1fb7700d073a37826e15dc697bd170816c8b Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Thu, 1 Oct 2026 06:34:59 +0800
Subject: [PATCH 1/7] refactor(aidd-refine): simplify improve instructions

---
 .../aidd-refine/skills/05-improve/SKILL.md    | 17 ++----
 .../actions/01-read-conversation.md           | 29 ++++------
 .../skills/05-improve/actions/02-recommend.md | 57 ++++++++-----------
 .../05-improve/actions/03-target-edits.md     | 28 ++++-----
 4 files changed, 54 insertions(+), 77 deletions(-)

diff --git a/plugins/aidd-refine/skills/05-improve/SKILL.md b/plugins/aidd-refine/skills/05-improve/SKILL.md
index 61f51420e..d2da36217 100644
--- a/plugins/aidd-refine/skills/05-improve/SKILL.md
+++ b/plugins/aidd-refine/skills/05-improve/SKILL.md
@@ -8,18 +8,14 @@ argument-hint: conversation | export
 ```mermaid
 flowchart LR
   start([conversation ID or export]) --> read-conversation
-  read-conversation -->|complete| scopes{two or more scopes?}
+  read-conversation -->|complete| recommend --> target-edits
   read-conversation -->|unavailable| unavailable([stop])
-  scopes -->|no| recommend-local[recommend locally] --> target-edits
-  scopes -->|yes| isolation{isolated artifact context?}
-  isolation -->|yes| recommend-parallel[recommend in parallel] --> target-edits
-  isolation -->|no| recommend-local
   target-edits --> question([ask next intent]) --> stop([stop])
 ```
 
 ## Actions
 
-Run the flow above. Read only the next action file.
+Run all three actions without confirmation. Read only the next action file.
 
 | Action | Does |
 | --- | --- |
@@ -29,8 +25,7 @@ Run the flow above. Read only the next action file.
 
 ## Transversal rules
 
-- After resolving the source, run all three actions without pausing for confirmation.
-- Analyze only a complete conversation and the exact skills or documents it names.
-- Exclude every current or previous `improve` invocation and report from the analysis evidence.
-- Never invent time, tokens, source coverage, or document status.
-- Do not modify, stage, or persist any project file.
+- Stop if the exact complete transcript is unavailable.
+- Assess only named artifacts alongside conversation behavior.
+- Never invent metrics, coverage, status, or private reasoning.
+- Keep the project read-only; write only the unique temporary report.
diff --git a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
index 1b146ce3a..a45802b7e 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
@@ -1,6 +1,6 @@
 # 01 - Read conversation
 
-Load complete conversation evidence once and profile its visible cost.
+Freeze complete evidence and measure visible cost.
 
 ## Input
 
@@ -12,24 +12,19 @@ An evidence boundary, a `## Timing` table with `Activity | Observed time | Share
 
 ## Process
 
-1. **Resolve.** Choose the host-specific transcript source from [conversation sources](../assets/conversation-sources.md).
-   - Stop when no complete transcript can be resolved for the exact conversation.
-2. **Bound.** Freeze the evidence at the message before the current invocation, or at the supplied export boundary.
-3. **Read.** Load every in-scope message, tool call, tool result, and timestamp from that transcript.
-4. **Index.** Record relevant turns and invoked or named skills and knowledge files for downstream analysis.
-5. **Measure.** Calculate visible time and usage only from timestamps, elapsed records, or host usage data.
-   - Group explicit tool activity as `research and diagnosis`, `implementation`, or `validation`; keep mixed or unknown time `unattributed`.
-   - Include tokens, requests, and monetary cost only when the source exposes them.
-   - Never label unattributed wait as reasoning time.
-6. **Render.** Mark a missing metric `unavailable` and order known activity times from longest to shortest.
+1. **Resolve.** Use the host route in [conversation sources](../assets/conversation-sources.md); stop unless it yields the exact complete transcript.
+2. **Freeze.** End before this invocation, or at the export boundary. Exclude every `improve` invocation and report.
+3. **Read once.** Load all in-scope messages, tool calls, results, and timestamps. Index relevant turns and named or invoked skill and knowledge paths.
+4. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
+   - Group tool time as `research and diagnosis`, `implementation`, `validation`, or `unattributed`.
+   - Include tokens, requests, and cost only when exposed.
+5. **Render.** Order known activity times descending; mark missing metrics `unavailable`. Never call unattributed or unavailable time private reasoning.
 
 ## Test
 
 | Case | Pass |
 | --- | --- |
-| A conversation is analyzed | its source resolves to one exact complete transcript |
-| The same conversation is analyzed again | its evidence excludes every `improve` invocation and report |
-| A timing metric is shown | its evidence identifies visible timestamp or elapsed records |
-| A metric is unavailable | the report does not estimate or call it reasoning time |
-| A usage metric is shown | its evidence identifies the host record that exposes it |
-| A scope is indexed | it identifies exact turns or artifact paths, not a generated summary |
+| Conversation | exact complete transcript; frozen boundary; no `improve` evidence |
+| Metric | cited host record, timestamp, or elapsed record; otherwise `unavailable` |
+| Time | visible categories only; no private-reasoning claim |
+| Scope | exact turns or artifact paths, never a generated summary |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
index 6e8f2681f..02de4fe1b 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
@@ -1,6 +1,6 @@
 # 02 - Recommend
 
-Analyze each relevant scope with the smallest grounded change.
+Find the smallest grounded improvements.
 
 ## Input
 
@@ -12,39 +12,32 @@ A `## Recommendations` table with `ID | Question | Type | Diagnostic | Evidence
 
 ## Process
 
-1. **Scope.** Select `behavior`, plus `skill` and `knowledge` only when the scope index names relevant artifacts.
-2. **Dispatch.** Analyze locally when one scope exists; otherwise dispatch one read-only analyst per scope in parallel.
-   - Prefer a lightweight available model and low reasoning effort when the host supports per-agent overrides; otherwise inherit the run defaults.
-   - Give the behavior analyst the complete frozen transcript; give artifact analysts the same boundary, indexed turns, and exact artifact paths.
-   - Request isolated or minimal context for artifact analysts when supported; otherwise analyze every scope locally instead of duplicating the transcript.
-   - Do not dispatch an analyst with no relevant evidence.
-   - Require table rows only and forbid file writes.
-3. **Question.** Make each analyst answer every prompt for its scope.
-   - How could the next run be faster or better?
-   - What information should be removed or clarified?
-   - Where should the change live?
-   - How could it save time or tokens?
-   - What work was counterproductive?
-4. **Verify.** Read a named skill or knowledge file before assessing its information.
-5. **Assess.** Label relevant information `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
-   - Use `correct` when no evidence supports another label, and never render it as a recommendation.
-6. **Merge.** Deduplicate findings across scopes and verify only their cited evidence against the frozen source.
-7. **Render.** Order by question then `behavior`, `skill`, `knowledge`, and describe each change with the fewest unambiguous words.
-   - Use `skill`, `behavior`, `knowledge`, or `tooling` as the target type.
-   - State `time`, `tokens`, `both`, or `unknown` as its saving.
-   - Render `no change` when a scope has no evidence-backed recommendation.
+1. **Scope.** Always assess `behavior`; add `skill` or `knowledge` only for indexed artifacts.
+2. **Dispatch.** With multiple scopes and isolated or minimal context, run one read-only analyst per scope in parallel. Otherwise analyze locally.
+   - Prefer a cheap model and low reasoning effort when per-agent overrides are supported; otherwise inherit host defaults.
+   - Give the behavior analyst the frozen transcript; give each artifact analyst the boundary, indexed turns, and paths.
+   - Give every analyst steps 3–7; require table rows and no writes.
+3. **Question.** Answer all for each scope:
+   - Faster or better next run? What was removable, unclear, or counterproductive?
+   - Where should change live? Would it save `time`, `tokens`, `both`, or `unknown`?
+   - Which back-and-forth, bottlenecks, or tool calls could be removed, batched, parallelized, or replaced?
+4. **Verify.** Read each named artifact before judging it.
+5. **Assess.** Use `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
+   - For skills, especially knowledge: delete evidence-backed waste or inconsistency first, without quota; then consolidate or clarify.
+   - Add only for a demonstrated gap the existing content cannot cover. Preserve useful context and requirements.
+   - `correct` means `no change`, never a recommendation.
+6. **Merge.** Deduplicate, then verify cited evidence against the frozen source.
+7. **Render.** Order by question, then `behavior`, `skill`, `knowledge`. Use the fewest actionable words.
+   - Type: `skill`, `behavior`, `knowledge`, or `tooling`.
+   - Saving: `time`, `tokens`, `both`, or `unknown`.
+   - Use `no change` when evidence supports none.
 
 ## Test
 
 | Case | Pass |
 | --- | --- |
-| A question is shown | it is answered from conversation evidence |
-| One relevant scope exists | no parallel analyst is dispatched |
-| Several relevant scopes exist | their analysts run in parallel and return the same columns |
-| An artifact analyst cannot receive isolated context | every scope is analyzed locally instead of duplicating the transcript |
-| A type is shown | it is `skill`, `behavior`, `knowledge`, or `tooling` |
-| Information is assessed | it has one allowed label backed by evidence |
-| Information is correct | its scope says `no change` and no recommendation is rendered |
-| A named target is assessed | the target file was read before the verdict |
-| A saving is shown | it is categorical and never an invented amount |
-| The same evidence is analyzed again | recommendations keep the same order and do not cite an earlier `improve` report |
+| Dispatch | parallel only for multiple isolated scopes; otherwise local |
+| Analysis | every question answered with exact evidence or `no change` |
+| Finding | exact turn, tool-call, or read-artifact evidence |
+| No finding | `no change`; no false recommendation |
+| Output | allowed labels, types, savings, and stable order; no `improve` evidence |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
index 53bc43068..71fffdf66 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
@@ -1,6 +1,6 @@
 # 03 - Target edits
 
-Map recommendations to minimal edits and render the report.
+Map minimal edits and render the report.
 
 ## Input
 
@@ -12,24 +12,18 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 
 ## Process
 
-1. **Target.** Map each file-targeted recommendation to a real project path.
-   - Omit behavior-only recommendations from this table.
-2. **Summarize.** Build a `Fichier | + Ajout | − Retrait ou clarification` table.
-   - Consolidate repeated targets and render each diagnostic's smallest change.
-   - Render one `Aucun fichier recommandé | — | —` row when no recommendation needs a file edit.
-3. **Render.** Fill [the report template](../assets/report-template.html) with only measured values and grounded findings, then copy its local CSS and JavaScript beside it.
-   - Remove every sample value and sample finding from the produced report.
-   - HTML-escape every injected value; allow only template-owned markup and local asset references.
-   - Write only to a unique temporary directory, never the project.
-4. **Return.** Provide the report path and end with this exact question: `Quel changement d’intention général, même minime, appliquons-nous au prochain run pour rendre notre amélioration cumulative et mesurable ?`
+1. **Target.** Map file recommendations to real paths; omit behavior-only rows. Build `Fichier | + Ajout | − Retrait ou clarification`, consolidate targets, and show `Aucun fichier recommandé | — | —` when empty.
+2. **Render.** Fill [the report template](../assets/report-template.html) with measured values and grounded findings; copy its local CSS and JavaScript.
+   - Remove samples. Escape every injected value; allow only template markup and local assets.
+   - Keep `data-prompt` and editable execution instructions as short as possible without losing targets or actions.
+   - Write only to the allowed unique temporary directory.
+3. **Return.** Provide its path and end exactly: `Quel changement d’intention général, même minime, appliquons-nous au prochain run pour rendre notre amélioration cumulative et mesurable ?`
 
 ## Test
 
 | Case | Pass |
 | --- | --- |
-| A file is shown | it is the target of an earlier recommendation |
-| No file needs editing | the empty-table row is shown |
-| The report is rendered | its HTML, CSS, and JavaScript resolve locally with no network dependency |
-| Conversation evidence is rendered | it is escaped as text and cannot create markup, scripts, or URLs |
-| A value is unavailable | the report says `unavailable` instead of showing sample data |
-| The run closes | the returned chat response ends with the exact next-intent question |
+| Edit table | prior file targets only; consolidated or empty row |
+| Report | no samples; unavailable values named; injected evidence inert |
+| Assets | HTML, CSS, and JavaScript resolve locally only |
+| Close | path returned; exact final question |

From 06347d51f5e28c195320fcc4a9c6c7f346aa1bc0 Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Thu, 1 Oct 2026 06:47:40 +0800
Subject: [PATCH 2/7] fix(aidd-refine): keep improve self-assessment internal

---
 .../skills/05-improve/actions/02-recommend.md       | 13 +++++++------
 .../skills/05-improve/actions/03-target-edits.md    |  6 +++---
 .../skills/05-improve/assets/report-template.html   |  4 ++--
 3 files changed, 12 insertions(+), 11 deletions(-)

diff --git a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
index 02de4fe1b..e57fea062 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
@@ -8,7 +8,7 @@ The evidence boundary, timing and usage tables, complete conversation, and scope
 
 ## Output
 
-A `## Recommendations` table with `ID | Question | Type | Diagnostic | Evidence | Smallest change | Target | Saving`.
+A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Target | Saving`.
 
 ## Process
 
@@ -17,9 +17,10 @@ A `## Recommendations` table with `ID | Question | Type | Diagnostic | Evidence
    - Prefer a cheap model and low reasoning effort when per-agent overrides are supported; otherwise inherit host defaults.
    - Give the behavior analyst the frozen transcript; give each artifact analyst the boundary, indexed turns, and paths.
    - Give every analyst steps 3–7; require table rows and no writes.
-3. **Question.** Answer all for each scope:
-   - Faster or better next run? What was removable, unclear, or counterproductive?
-   - Where should change live? Would it save `time`, `tokens`, `both`, or `unknown`?
+3. **Reflect.** Ask yourself, for each scope; keep this assessment internal:
+   - How could the next run be faster or better?
+   - What should be removed or clarified? What was counterproductive?
+   - Where should change live? How could it save time or tokens?
    - Which back-and-forth, bottlenecks, or tool calls could be removed, batched, parallelized, or replaced?
 4. **Verify.** Read each named artifact before judging it.
 5. **Assess.** Use `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
@@ -27,7 +28,7 @@ A `## Recommendations` table with `ID | Question | Type | Diagnostic | Evidence
    - Add only for a demonstrated gap the existing content cannot cover. Preserve useful context and requirements.
    - `correct` means `no change`, never a recommendation.
 6. **Merge.** Deduplicate, then verify cited evidence against the frozen source.
-7. **Render.** Order by question, then `behavior`, `skill`, `knowledge`. Use the fewest actionable words.
+7. **Render.** Order by focus, then `behavior`, `skill`, `knowledge`. Report findings, not the questionnaire. Use the fewest actionable words.
    - Type: `skill`, `behavior`, `knowledge`, or `tooling`.
    - Saving: `time`, `tokens`, `both`, or `unknown`.
    - Use `no change` when evidence supports none.
@@ -37,7 +38,7 @@ A `## Recommendations` table with `ID | Question | Type | Diagnostic | Evidence
 | Case | Pass |
 | --- | --- |
 | Dispatch | parallel only for multiple isolated scopes; otherwise local |
-| Analysis | every question answered with exact evidence or `no change` |
+| Analysis | internal assessment answers every question with exact evidence or `no change` |
 | Finding | exact turn, tool-call, or read-artifact evidence |
 | No finding | `no change`; no false recommendation |
 | Output | allowed labels, types, savings, and stable order; no `improve` evidence |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
index 71fffdf66..cc170257d 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
@@ -8,7 +8,7 @@ The timing, usage, and recommendations tables.
 
 ## Output
 
-An HTML report in a temporary directory with local `report.css` and `report.js`, plus its path and the next-intent question.
+An HTML report in a temporary directory with local `report.css` and `report.js`, plus its path and a next-intent request.
 
 ## Process
 
@@ -17,7 +17,7 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
    - Remove samples. Escape every injected value; allow only template markup and local assets.
    - Keep `data-prompt` and editable execution instructions as short as possible without losing targets or actions.
    - Write only to the allowed unique temporary directory.
-3. **Return.** Provide its path and end exactly: `Quel changement d’intention général, même minime, appliquons-nous au prochain run pour rendre notre amélioration cumulative et mesurable ?`
+3. **Return.** Return its path. Ask the user, in their language, which small general change in intent to apply next run for cumulative, measurable improvement.
 
 ## Test
 
@@ -26,4 +26,4 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 | Edit table | prior file targets only; consolidated or empty row |
 | Report | no samples; unavailable values named; injected evidence inert |
 | Assets | HTML, CSS, and JavaScript resolve locally only |
-| Close | path returned; exact final question |
+| Close | path returned; next intent requested in the user's language |
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report-template.html b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
index 0c5300815..292c0078e 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report-template.html
+++ b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
@@ -3,7 +3,7 @@
   Template contract: replace every sample run value, timing, finding and file row
   with evidence from the current conversation. Duplicate or remove `.finding`
   blocks as needed, keep their data-signal/data-target values aligned with the
-  visible tags, derive the compact question tag from the finding question, write
+  visible tags, derive the compact focus tag from the finding focus, write
   each accepted action in its data-prompt attribute, and ship report.css/report.js
   beside this file. Remove this comment from the report.
 -->
@@ -198,7 +198,7 @@ <h4>Preuve</h4>
 
 Changements acceptés :
 
-Une fois le travail validé, pose-moi cette question : « Quel changement d’intention général, même minime, appliquons-nous au prochain run pour rendre notre amélioration cumulative et mesurable ? »</textarea>
+Après validation, propose un changement général pour améliorer le prochain run et mesurer son effet.</textarea>
         <div class="prompt-actions">
           <span>Le texte reste éditable.</span>
           <button class="button" id="copy-prompt" type="button">Copier le prompt</button>

From 3147994843b027534a054aed58a44bf93ef0a635 Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Thu, 1 Oct 2026 06:50:26 +0800
Subject: [PATCH 3/7] fix(aidd-refine): keep next-run intent change small

---
 .../aidd-refine/skills/05-improve/assets/report-template.html   | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/plugins/aidd-refine/skills/05-improve/assets/report-template.html b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
index 292c0078e..3ac50ad60 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report-template.html
+++ b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
@@ -198,7 +198,7 @@ <h4>Preuve</h4>
 
 Changements acceptés :
 
-Après validation, propose un changement général pour améliorer le prochain run et mesurer son effet.</textarea>
+Après validation, propose un petit changement général pour améliorer le prochain run et mesurer son effet.</textarea>
         <div class="prompt-actions">
           <span>Le texte reste éditable.</span>
           <button class="button" id="copy-prompt" type="button">Copier le prompt</button>

From 79660f07580c24532bc65a9607013ea54d1e600e Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Mon, 5 Oct 2026 08:55:50 +0700
Subject: [PATCH 4/7] fix(aidd-refine): make improve recommendations actionable
 and efficient

---
 .../aidd-refine/skills/05-improve/SKILL.md    |   2 +-
 .../actions/01-read-conversation.md           |  14 +-
 .../skills/05-improve/actions/02-recommend.md |  21 +-
 .../05-improve/actions/03-target-edits.md     |  12 +-
 .../05-improve/assets/report-template.html    | 214 +++---------
 .../skills/05-improve/assets/report.css       | 328 ++++--------------
 .../skills/05-improve/assets/report.js        |  51 ++-
 scripts/__tests__/improve-report.test.js      |  46 +++
 8 files changed, 242 insertions(+), 446 deletions(-)
 create mode 100644 scripts/__tests__/improve-report.test.js

diff --git a/plugins/aidd-refine/skills/05-improve/SKILL.md b/plugins/aidd-refine/skills/05-improve/SKILL.md
index d2da36217..509d26ec2 100644
--- a/plugins/aidd-refine/skills/05-improve/SKILL.md
+++ b/plugins/aidd-refine/skills/05-improve/SKILL.md
@@ -26,6 +26,6 @@ Run all three actions without confirmation. Read only the next action file.
 ## Transversal rules
 
 - Stop if the exact complete transcript is unavailable.
-- Assess only named artifacts alongside conversation behavior.
+- Assess conversation behavior, invoked skill resources, applicable project instructions, and task-relevant indexed memory.
 - Never invent metrics, coverage, status, or private reasoning.
 - Keep the project read-only; write only the unique temporary report.
diff --git a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
index a45802b7e..d23d1d1e6 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
@@ -8,17 +8,19 @@ A current conversation, an exact conversation ID, or a complete transcript expor
 
 ## Output
 
-An evidence boundary, a `## Timing` table with `Activity | Observed time | Share | Evidence`, a `## Usage` table with `Metric | Value | Evidence`, and a scope index.
+An evidence boundary, a `## Timing` table with `Activity | Observed time | Share | Evidence`, a `## Usage` table with `Metric | Value | Evidence`, and a `Scope | Status | Evidence` index.
 
 ## Process
 
 1. **Resolve.** Use the host route in [conversation sources](../assets/conversation-sources.md); stop unless it yields the exact complete transcript.
 2. **Freeze.** End before this invocation, or at the export boundary. Exclude every `improve` invocation and report.
-3. **Read once.** Load all in-scope messages, tool calls, results, and timestamps. Index relevant turns and named or invoked skill and knowledge paths.
-4. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
+3. **Read once.** Load all in-scope messages, tool calls, results, and timestamps. Index relevant turns, invoked skills with used actions or resources, and applicable project instructions.
+4. **Context.** Read indexed skill resources and applicable `AGENTS.md`. Use the project memory index to read only task-relevant memory; never scan the whole memory library. Mark only read files `checked`; mark unresolved or skipped scopes `missing` or `not reviewed`.
+5. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
    - Group tool time as `research and diagnosis`, `implementation`, `validation`, or `unattributed`.
    - Include tokens, requests, and cost only when exposed.
-5. **Render.** Order known activity times descending; mark missing metrics `unavailable`. Never call unattributed or unavailable time private reasoning.
+   - Separate background-process lifetime from blocking time.
+6. **Render.** Order known activity times descending; mark missing metrics `unavailable`. Never call unattributed or unavailable time private reasoning.
 
 ## Test
 
@@ -26,5 +28,5 @@ An evidence boundary, a `## Timing` table with `Activity | Observed time | Share
 | --- | --- |
 | Conversation | exact complete transcript; frozen boundary; no `improve` evidence |
 | Metric | cited host record, timestamp, or elapsed record; otherwise `unavailable` |
-| Time | visible categories only; no private-reasoning claim |
-| Scope | exact turns or artifact paths, never a generated summary |
+| Time | visible categories only; background lifetime separated; no private-reasoning claim |
+| Scope | exact turns or paths; only read files are `checked` |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
index e57fea062..10a5e5319 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
@@ -1,6 +1,6 @@
 # 02 - Recommend
 
-Find the smallest grounded improvements.
+Reduce avoidable work without reducing reliability.
 
 ## Input
 
@@ -8,11 +8,11 @@ The evidence boundary, timing and usage tables, complete conversation, and scope
 
 ## Output
 
-A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Target | Saving`.
+A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Source | Saving`, plus non-actionable limitations.
 
 ## Process
 
-1. **Scope.** Always assess `behavior`; add `skill` or `knowledge` only for indexed artifacts.
+1. **Scope.** Always assess `behavior`; add `skill` or `knowledge` for indexed skill resources, instructions, or memory.
 2. **Dispatch.** With multiple scopes and isolated or minimal context, run one read-only analyst per scope in parallel. Otherwise analyze locally.
    - Prefer a cheap model and low reasoning effort when per-agent overrides are supported; otherwise inherit host defaults.
    - Give the behavior analyst the frozen transcript; give each artifact analyst the boundary, indexed turns, and paths.
@@ -22,13 +22,19 @@ A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | S
    - What should be removed or clarified? What was counterproductive?
    - Where should change live? How could it save time or tokens?
    - Which back-and-forth, bottlenecks, or tool calls could be removed, batched, parallelized, or replaced?
-4. **Verify.** Read each named artifact before judging it.
+4. **Verify.** Resolve and read each maintained source before judging it; never target caches or installs. Account for intentional host transforms. Judge only `checked` sources; an installed artifact never makes its source `checked`.
 5. **Assess.** Use `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
+   - Compare observed work with the shortest reliable path to the requested result. Keep checks for mutable state or unresolved uncertainty; reuse still-valid results instead of repeating research.
+   - Raw call count never proves waste. Ground the smallest general correction in evidence.
+   - Recommend memory only for concise, durable knowledge that prevents recurring rediscovery, never transient state or one-off implementation details.
    - For skills, especially knowledge: delete evidence-backed waste or inconsistency first, without quota; then consolidate or clarify.
-   - Add only for a demonstrated gap the existing content cannot cover. Preserve useful context and requirements.
+   - Reuse or strengthen existing content; add only for a demonstrated gap it cannot cover. Preserve useful context and requirements; use `correct` when it already suffices.
+   - Claim time savings from blocking-time or resource-impact evidence, never background-process uptime alone.
    - `correct` means `no change`, never a recommendation.
 6. **Merge.** Deduplicate, then verify cited evidence against the frozen source.
 7. **Render.** Order by focus, then `behavior`, `skill`, `knowledge`. Report findings, not the questionnaire. Use the fewest actionable words.
+   - Every recommendation, including behavior or tooling, needs an exact maintained source path, exact current excerpt, minimal replacement or explicit `+`/`−` edit, and a short target/action prompt.
+   - If no source resolves or no useful persistent edit exists, render a collapsed limitation, not an accept-toggle recommendation.
    - Type: `skill`, `behavior`, `knowledge`, or `tooling`.
    - Saving: `time`, `tokens`, `both`, or `unknown`.
    - Use `no change` when evidence supports none.
@@ -39,6 +45,9 @@ A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | S
 | --- | --- |
 | Dispatch | parallel only for multiple isolated scopes; otherwise local |
 | Analysis | internal assessment answers every question with exact evidence or `no change` |
-| Finding | exact turn, tool-call, or read-artifact evidence |
+| Efficiency | shortest reliable path; necessary checks retained; repeated work evidenced |
+| Finding | exact evidence, maintained source path, minimal edit, and target/action prompt |
 | No finding | `no change`; no false recommendation |
+| Coverage | verdicts only for `checked` files; other scopes stay explicit |
+| Limitation | unresolved or non-persistent work is not actionable |
 | Output | allowed labels, types, savings, and stable order; no `improve` evidence |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
index cc170257d..ddc8cc87a 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
@@ -4,7 +4,7 @@ Map minimal edits and render the report.
 
 ## Input
 
-The timing, usage, and recommendations tables.
+The timing, usage, recommendations, and scope index tables.
 
 ## Output
 
@@ -12,9 +12,12 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 
 ## Process
 
-1. **Target.** Map file recommendations to real paths; omit behavior-only rows. Build `Fichier | + Ajout | − Retrait ou clarification`, consolidate targets, and show `Aucun fichier recommandé | — | —` when empty.
+1. **Target.** Map every recommendation to its maintained source path. Build `Fichier | + Ajout | − Retrait ou clarification`, consolidate targets, and show `Aucun fichier recommandé | — | —` when empty. Recommend only; never apply.
 2. **Render.** Fill [the report template](../assets/report-template.html) with measured values and grounded findings; copy its local CSS and JavaScript.
    - Remove samples. Escape every injected value; allow only template markup and local assets.
+   - Put findings first. Preserve diagnostic/type tags, collapsible filters, accept controls, target table, and prompt.
+   - Show each source path and minimal edit. Keep raw evidence, secondary metrics, timing, coverage, source records, and limitations collapsed.
+   - Remove repeated metadata, focus/action tags, score labels, verbose help, and redundant run metadata; put the run ID in collapsed source records.
    - Keep `data-prompt` and editable execution instructions as short as possible without losing targets or actions.
    - Write only to the allowed unique temporary directory.
 3. **Return.** Return its path. Ask the user, in their language, which small general change in intent to apply next run for cumulative, measurable improvement.
@@ -23,7 +26,8 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 
 | Case | Pass |
 | --- | --- |
-| Edit table | prior file targets only; consolidated or empty row |
-| Report | no samples; unavailable values named; injected evidence inert |
+| Edit table | every actionable finding has a maintained source row; consolidated or empty |
+| Report | no samples; unavailable values and scope coverage named; injected evidence inert |
 | Assets | HTML, CSS, and JavaScript resolve locally only |
+| Findings | exact source edit and accept control; evidence and limitations collapse |
 | Close | path returned; next intent requested in the user's language |
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report-template.html b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
index 3ac50ad60..0d0f26cae 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report-template.html
+++ b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
@@ -1,11 +1,9 @@
 <!doctype html>
 <!--
-  Template contract: replace every sample run value, timing, finding and file row
-  with evidence from the current conversation. Duplicate or remove `.finding`
-  blocks as needed, keep their data-signal/data-target values aligned with the
-  visible tags, derive the compact focus tag from the finding focus, write
-  each accepted action in its data-prompt attribute, and ship report.css/report.js
-  beside this file. Remove this comment from the report.
+  Replace all samples with current evidence. Each actionable `.finding` needs
+  diagnostic/type tags, a maintained source path, minimal edit, evidence, and
+  target/action data-prompt. Put unresolved or non-persistent items in collapsed
+  limitations. Keep report.css/report.js local. Remove this comment in output.
 -->
 <html lang="fr">
   <head>
@@ -18,191 +16,89 @@
   </head>
   <body>
     <header class="site-header">
-      <a class="brand" href="#report" aria-label="AIDD Improve report">
-        <span class="brand-mark" aria-hidden="true">A</span>
-        <span>AIDD / improve</span>
-      </a>
-      <div class="header-actions">
-        <span class="run-id">run_2026-09-12_01</span>
-        <button class="button" id="print-report" type="button">Imprimer</button>
-      </div>
+      <a class="brand" href="#report" aria-label="AIDD Improve report"><span class="brand-mark" aria-hidden="true">A</span><span>AIDD / improve</span></a>
+      <div class="header-actions"><button class="button" id="print-report" type="button">Imprimer</button></div>
     </header>
 
     <main id="report">
       <section class="intro" aria-labelledby="report-title">
-        <p class="context">Conversation actuelle · couverture partielle</p>
         <h1 id="report-title">Rapport d’amélioration</h1>
-        <p class="lede">
-          Les changements les plus courts qui rendront le prochain run plus rapide,
-          plus fiable et moins coûteux.
-        </p>
-        <div class="run-meta">
-          <p><span>Généré</span><strong>12 septembre 2026 · 10:42</strong></p>
-          <p><span>Source</span><strong>Conversation + fichiers cités</strong></p>
-          <p><span>Findings</span><strong>3 recommandations</strong></p>
-        </div>
-      </section>
-
-      <section class="section" aria-labelledby="cost-title">
-        <div class="section-heading">
-          <div><p class="section-index">Mesure</p><h2 id="cost-title">Coût de la session</h2></div>
-          <span class="coverage">Données partielles</span>
-        </div>
-        <div class="metrics" role="list" aria-label="Mesures de la session">
-          <div class="metric" role="listitem"><span>Durée active</span><strong>24 min</strong><small>mesurée</small></div>
-          <div class="metric" role="listitem"><span>Tokens</span><strong>128,4 k</strong><small>entrée + sortie + cache</small></div>
-          <div class="metric" role="listitem"><span>Requêtes</span><strong>42</strong><small>tous outils</small></div>
-          <div class="metric metric-unknown" role="listitem"><span>Coût monétaire</span><strong>Inconnu</strong><small>non exposé par le fournisseur</small></div>
-        </div>
-        <p class="source-note">Source : <code>aidd telemetry report v8</code> · 2026-09-12T02:00Z → 2026-09-12T03:00Z</p>
-      </section>
-
-      <section class="section" aria-labelledby="timing-title">
-        <div class="section-heading">
-          <div><p class="section-index">Chronologie</p><h2 id="timing-title">Où le temps est parti</h2></div>
-          <span class="section-total">24 min au total</span>
-        </div>
-        <ol class="timing-list">
-          <li><span class="timing-rank">01</span><div><strong>Recherche et diagnostic</strong><span>12 min · 50 %</span></div><progress class="timing-bar" max="100" value="100">100 %</progress></li>
-          <li><span class="timing-rank">02</span><div><strong>Implémentation</strong><span>8 min · 33 %</span></div><progress class="timing-bar" max="100" value="66">66 %</progress></li>
-          <li><span class="timing-rank">03</span><div><strong>Validation</strong><span>4 min · 17 %</span></div><progress class="timing-bar" max="100" value="33">33 %</progress></li>
-        </ol>
+        <p>Exemple · conversation actuelle · couverture partielle · 1 recommandation</p>
       </section>
 
       <section class="section findings-section" aria-labelledby="findings-title">
-        <div class="section-heading findings-heading">
-          <div><p class="section-index">Décisions</p><h2 id="findings-title">Changements recommandés</h2></div>
-          <p id="result-count" class="result-count" aria-live="polite">3 recommandations</p>
-        </div>
+        <div class="section-heading"><h2 id="findings-title">Changements recommandés</h2><p id="result-count" class="result-count" aria-live="polite">1 recommandation</p></div>
 
-        <div class="filters" aria-label="Filtrer les recommandations">
-          <div class="filter-group" role="group" aria-label="Diagnostic">
-            <span>Diagnostic</span>
-            <button type="button" data-filter="signal" data-value="all" aria-pressed="true">Tous</button>
-            <button type="button" data-filter="signal" data-value="obsolete" aria-pressed="false">Obsolète</button>
-            <button type="button" data-filter="signal" data-value="over-specific-or-time-bound" aria-pressed="false">Trop spécifique ou temporel</button>
-            <button type="button" data-filter="signal" data-value="duplicate" aria-pressed="false">Doublon</button>
-            <button type="button" data-filter="signal" data-value="inconsistent" aria-pressed="false">Incohérent</button>
-            <button type="button" data-filter="signal" data-value="counterproductive" aria-pressed="false">Contre-productif</button>
-          </div>
-          <div class="filter-group" role="group" aria-label="Type de cible">
-            <span>Cible</span>
-            <button type="button" data-filter="target" data-value="all" aria-pressed="true">Toutes</button>
-            <button type="button" data-filter="target" data-value="behavior" aria-pressed="false">Behavior</button>
-            <button type="button" data-filter="target" data-value="skill" aria-pressed="false">Skill</button>
-            <button type="button" data-filter="target" data-value="knowledge" aria-pressed="false">Knowledge</button>
-            <button type="button" data-filter="target" data-value="tooling" aria-pressed="false">Tooling</button>
+        <details class="filter-panel">
+          <summary>Filtres</summary>
+          <div class="filters" aria-label="Filtrer les recommandations">
+            <div class="filter-group" role="group" aria-label="Diagnostic">
+              <span>Diagnostic</span>
+              <button type="button" data-filter="signal" data-value="all" aria-pressed="true">Tous</button>
+              <button type="button" data-filter="signal" data-value="obsolete" aria-pressed="false">Obsolète</button>
+              <button type="button" data-filter="signal" data-value="over-specific-or-time-bound" aria-pressed="false">Trop spécifique</button>
+              <button type="button" data-filter="signal" data-value="duplicate" aria-pressed="false">Doublon</button>
+              <button type="button" data-filter="signal" data-value="inconsistent" aria-pressed="false">Incohérent</button>
+              <button type="button" data-filter="signal" data-value="counterproductive" aria-pressed="false">Contre-productif</button>
+            </div>
+            <div class="filter-group" role="group" aria-label="Type de cible">
+              <span>Cible</span>
+              <button type="button" data-filter="target" data-value="all" aria-pressed="true">Toutes</button>
+              <button type="button" data-filter="target" data-value="behavior" aria-pressed="false">Behavior</button>
+              <button type="button" data-filter="target" data-value="skill" aria-pressed="false">Skill</button>
+              <button type="button" data-filter="target" data-value="knowledge" aria-pressed="false">Knowledge</button>
+              <button type="button" data-filter="target" data-value="tooling" aria-pressed="false">Tooling</button>
+            </div>
           </div>
-        </div>
+        </details>
 
         <div class="findings" id="findings-list">
-          <article class="finding" data-signal="obsolete" data-target="tooling">
+          <article class="finding" data-signal="inconsistent" data-target="skill">
             <header class="finding-header">
               <div class="finding-id">F-001</div>
-              <div class="finding-title">
-                <div class="tags"><span class="tag tag-question">Plus rapide</span><span class="tag tag-signal">Obsolète</span><span class="tag">Tooling</span><span class="tag tag-action">Mettre à jour</span></div>
-                <h3>La version flottante casse le reload</h3>
-              </div>
-              <div class="finding-score"><p><span>Économie</span><strong>Temps</strong></p></div>
-              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="Épingler une version compatible dans scripts/dev-sync.sh afin que le reload reste reproductible."><span>Accepter</span></label>
+              <div class="finding-title"><div class="tags"><span class="tag tag-signal">Incohérent</span><span class="tag">Skill</span></div><h3>Exemple · préciser la preuve d’économie</h3></div>
+              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="Dans plugins/aidd-refine/skills/05-improve/actions/02-recommend.md, préciser que l’uptime seul ne prouve aucune économie."><span>Accepter</span></label>
             </header>
             <div class="finding-collapse"><div class="finding-body">
-              <div class="finding-copy">
-                <h4>Pourquoi</h4>
-                <p><code>latest</code> pointe désormais vers une CLI dont la commande <code>framework build</code> n’existe plus.</p>
-                <h4>Preuve</h4>
-                <p class="evidence">Commande observée : <code>error: unknown command 'build'</code></p>
-              </div>
               <div class="change">
-                <div class="change-header"><span>scripts/dev-sync.sh</span><button class="copy-path" type="button" data-copy="scripts/dev-sync.sh">Copier le chemin</button></div>
-                <pre class="diff" aria-label="Diff proposé pour scripts/dev-sync.sh"><code><span class="diff-line diff-del"><b>−</b>AIDD_CLI_VERSION="${AIDD_CLI_VERSION:-latest}"</span>
-<span class="diff-line diff-add"><b>+</b>AIDD_CLI_VERSION="${AIDD_CLI_VERSION:-5.2.2}"</span></code></pre>
+                <div class="change-header"><span>plugins/aidd-refine/skills/05-improve/actions/02-recommend.md</span><button class="copy-path" type="button" data-copy="plugins/aidd-refine/skills/05-improve/actions/02-recommend.md">Copier</button></div>
+                <pre class="diff" aria-label="Modification proposée pour 02-recommend.md"><code><span class="diff-line diff-del"><b>−</b>Claim time savings from blocking-time or resource-impact evidence, never background-process uptime alone.</span>
+<span class="diff-line diff-add"><b>+</b>Claim time savings only from blocking-time or resource-impact evidence; background uptime alone proves none.</span></code></pre>
               </div>
+              <p class="why">Sépare durée observée et impact réel.</p>
+              <details class="evidence-details"><summary>Preuve</summary><p>Le processus observé restait actif sans bloquer la suite.</p></details>
             </div></div>
           </article>
 
-          <article class="finding" data-signal="duplicate" data-target="skill">
-            <header class="finding-header">
-              <div class="finding-id">F-002</div>
-              <div class="finding-title">
-                <div class="tags"><span class="tag tag-question">Économiser des tokens</span><span class="tag tag-signal">Doublon</span><span class="tag">Skill</span><span class="tag tag-action">Fusionner</span></div>
-                <h3>Deux règles décrivent le même contrôle</h3>
-              </div>
-              <div class="finding-score"><p><span>Économie</span><strong>Tokens</strong></p></div>
-              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="Supprimer la règle dupliquée de skills/05-improve/SKILL.md et conserver l’action recommend comme source unique."><span>Accepter</span></label>
-            </header>
-            <div class="finding-collapse"><div class="finding-body">
-              <div class="finding-copy">
-                <h4>Pourquoi</h4>
-                <p>Le routeur et l’action répètent le même invariant. Une seule source suffit.</p>
-                <h4>Preuve</h4>
-                <p class="evidence">Même exigence relevée dans <code>SKILL.md:24</code> et <code>actions/02-recommend.md:31</code>.</p>
-              </div>
-              <div class="change">
-                <div class="change-header"><span>skills/05-improve/SKILL.md</span><button class="copy-path" type="button" data-copy="skills/05-improve/SKILL.md">Copier le chemin</button></div>
-                <pre class="diff" aria-label="Diff proposé pour SKILL.md"><code><span class="diff-line diff-del"><b>−</b>Verify every recommendation against evidence.</span>
-<span class="diff-line diff-context"><b> </b>The recommend action owns evidence verification.</span></code></pre>
-              </div>
-            </div></div>
-          </article>
-
-          <article class="finding" data-signal="inconsistent" data-target="knowledge">
-            <header class="finding-header">
-              <div class="finding-id">F-003</div>
-              <div class="finding-title">
-                <div class="tags"><span class="tag tag-question">Supprimer ou clarifier</span><span class="tag tag-signal">Incohérent</span><span class="tag">Knowledge</span><span class="tag tag-action">Clarifier</span></div>
-                <h3>La documentation promet un coût indisponible</h3>
-              </div>
-              <div class="finding-score"><p><span>Économie</span><strong>Temps</strong></p></div>
-              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="Clarifier dans aidd_docs/memory/telemetry.md que le coût est affiché seulement lorsqu’il est mesuré, sinon indiquer Inconnu."><span>Accepter</span></label>
-            </header>
-            <div class="finding-collapse"><div class="finding-body">
-              <div class="finding-copy">
-                <h4>Pourquoi</h4>
-                <p>Le montant exact n’est pas exposé. Le rapport doit distinguer valeur inconnue et zéro.</p>
-                <h4>Preuve</h4>
-                <p class="evidence"><code>cost_micro_usd</code> absent des observations de cette session.</p>
-              </div>
-              <div class="change">
-                <div class="change-header"><span>aidd_docs/memory/telemetry.md</span><button class="copy-path" type="button" data-copy="aidd_docs/memory/telemetry.md">Copier le chemin</button></div>
-                <pre class="diff" aria-label="Diff proposé pour telemetry.md"><code><span class="diff-line diff-del"><b>−</b>Le coût de chaque session est affiché.</span>
-<span class="diff-line diff-add"><b>+</b>Le coût est affiché lorsqu’il est mesuré ; sinon : Inconnu.</span></code></pre>
-              </div>
-            </div></div>
-          </article>
         </div>
-        <p class="empty-state" id="empty-state" hidden>Aucun finding ne correspond à ces filtres.</p>
+        <p class="empty-state" id="empty-state" hidden>Aucune recommandation ne correspond.</p>
+
+        <details class="limitations"><summary>Limites non actionnables · 1</summary><p>Une source citée n’a pas pu être résolue ; aucun changement n’est proposé.</p></details>
       </section>
 
       <section class="section target-section" aria-labelledby="target-title">
-        <div class="section-heading"><div><p class="section-index">Synthèse</p><h2 id="target-title">Fichiers ciblés</h2></div><span class="section-total">3 fichiers</span></div>
-        <div class="table-wrap">
-          <table>
-            <thead><tr><th>Fichier</th><th>+ Ajout</th><th>− Retrait ou clarification</th></tr></thead>
-            <tbody>
-              <tr><td><code>scripts/dev-sync.sh</code></td><td>Épingler une version compatible</td><td>Retirer <code>latest</code></td></tr>
-              <tr><td><code>skills/05-improve/SKILL.md</code></td><td>Conserver un propriétaire</td><td>Retirer la règle dupliquée</td></tr>
-              <tr><td><code>aidd_docs/memory/telemetry.md</code></td><td>Expliciter l’absence de mesure</td><td>Remplacer la promesse absolue</td></tr>
-            </tbody>
-          </table>
-        </div>
+        <div class="section-heading"><h2 id="target-title">Fichiers ciblés</h2><span class="section-total">1 fichier</span></div>
+        <div class="table-wrap"><table><thead><tr><th>Fichier</th><th>+ Ajout</th><th>− Retrait ou clarification</th></tr></thead><tbody>
+          <tr><td><code>plugins/aidd-refine/skills/05-improve/actions/02-recommend.md</code></td><td>« uptime seul = aucune économie prouvée »</td><td>Formulation ambiguë</td></tr>
+        </tbody></table></div>
       </section>
 
+      <details class="section raw-section">
+        <summary>Mesures, chronologie et couverture</summary>
+        <div class="metrics"><p><span>Durée active</span><strong>24 min</strong></p><p><span>Tokens</span><strong>128,4 k</strong></p><p><span>Requêtes</span><strong>42</strong></p><p><span>Coût</span><strong>Inconnu</strong></p></div>
+        <div class="table-wrap"><table><thead><tr><th>Activité</th><th>Temps</th><th>Part</th></tr></thead><tbody><tr><td>Recherche et diagnostic</td><td>12 min</td><td>50 %</td></tr><tr><td>Implémentation</td><td>8 min</td><td>33 %</td></tr><tr><td>Validation</td><td>4 min</td><td>17 %</td></tr></tbody></table></div>
+        <h3>Couverture</h3><p><code>AGENTS.md</code> · checked<br /><code>aidd_docs/memory/testing.md</code> · not reviewed<br /><code>skill export</code> · missing</p>
+        <details class="source-records"><summary>Sources brutes</summary><p><code>run_2026-09-12_01</code> · <code>aidd telemetry report v8</code> · 2026-09-12T02:00Z → 03:00Z</p></details>
+      </details>
+
       <footer class="prompt-builder" aria-labelledby="prompt-title">
-        <div class="prompt-heading">
-          <div><p>Passage à l’action</p><h2 id="prompt-title">Prompt d’exécution</h2></div>
-          <span id="accepted-count" aria-live="polite">0 recommandation acceptée</span>
-        </div>
-        <p class="prompt-help">Acceptez les changements à appliquer, ajustez le texte si nécessaire, puis copiez-le dans votre conversation avec l’IA.</p>
-        <textarea id="execution-prompt" rows="10" spellcheck="true">Applique uniquement les changements acceptés ci-dessous. Vérifie chaque cible avant modification, préserve les changements existants, puis valide le résultat.
+        <div class="prompt-heading"><h2 id="prompt-title">Prompt d’exécution</h2><span id="accepted-count" aria-live="polite">0 recommandation acceptée</span></div>
+        <textarea id="execution-prompt" rows="10" spellcheck="true">Applique uniquement les changements acceptés. Vérifie chaque source, préserve les changements existants, puis valide.
 
 Changements acceptés :
 
 Après validation, propose un petit changement général pour améliorer le prochain run et mesurer son effet.</textarea>
-        <div class="prompt-actions">
-          <span>Le texte reste éditable.</span>
-          <button class="button" id="copy-prompt" type="button">Copier le prompt</button>
-        </div>
+        <div class="prompt-actions"><span>Le texte reste éditable.</span><button class="button" id="copy-prompt" type="button">Copier le prompt</button></div>
       </footer>
     </main>
   </body>
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report.css b/plugins/aidd-refine/skills/05-improve/assets/report.css
index 1f5a54a28..b7da388d5 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report.css
+++ b/plugins/aidd-refine/skills/05-improve/assets/report.css
@@ -6,293 +6,117 @@
   --green-dark: color-mix(in oklch, var(--green) 62%, black);
   --green-soft: color-mix(in oklch, var(--green) 15%, white);
   --ink: oklch(0.16 0.01 25);
-  --ink-soft: oklch(0.38 0.01 25);
-  --muted: oklch(0.53 0.008 25);
+  --soft-ink: oklch(0.4 0.01 25);
+  --muted: oklch(0.55 0.008 25);
   --line: oklch(0.87 0.006 25);
-  --line-strong: oklch(0.72 0.01 25);
-  --surface: oklch(1 0 0);
-  --surface-soft: oklch(0.975 0.003 25);
-  --add-bg: var(--green-soft);
-  --add-ink: var(--green-dark);
-  --del-bg: var(--red-soft);
-  --del-ink: var(--red-dark);
+  --strong-line: oklch(0.72 0.01 25);
+  --surface: white;
+  --soft-surface: oklch(0.975 0.003 25);
   --focus: oklch(0.64 0.18 250);
   --mono: "SFMono-Regular", Consolas, "Liberation Mono", Menlo, monospace;
 }
 
 * { box-sizing: border-box; }
-
-html { scroll-behavior: smooth; }
-
-body {
-  margin: 0;
-  background: var(--surface);
-  color: var(--ink);
-  font-family: var(--mono);
-  font-size: 14px;
-  line-height: 1.6;
-}
-
-button, a { font: inherit; }
-
-button:focus-visible,
-a:focus-visible {
-  outline: 3px solid var(--focus);
-  outline-offset: 3px;
-}
-
+body { margin: 0; background: var(--surface); color: var(--ink); font: 14px/1.6 var(--mono); }
+button, textarea { font: inherit; }
+button:focus-visible, summary:focus-visible, textarea:focus { outline: 3px solid var(--focus); outline-offset: 3px; }
 code { font-family: var(--mono); }
 
-.site-header {
-  min-height: 58px;
-  padding: 0 32px;
-  border-bottom: 1px solid var(--line);
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  gap: 24px;
-  position: sticky;
-  top: 0;
-  z-index: 10;
-  background: color-mix(in oklch, var(--surface) 94%, transparent);
-  backdrop-filter: blur(12px);
-}
-
-.brand {
-  color: var(--ink);
-  text-decoration: none;
-  display: inline-flex;
-  align-items: center;
-  gap: 10px;
-  font-weight: 700;
-  letter-spacing: -0.02em;
-}
-
-.brand-mark {
-  width: 25px;
-  height: 25px;
-  display: grid;
-  place-items: center;
-  background: var(--red);
-  color: var(--surface);
-  font-size: 13px;
-}
-
-.header-actions { display: flex; align-items: center; gap: 16px; }
-.run-id { color: var(--muted); font-size: 12px; }
-
-.button {
-  border: 1px solid var(--ink);
-  background: var(--ink);
-  color: var(--surface);
-  padding: 7px 12px;
-  cursor: pointer;
-}
-
-.button:hover { background: var(--red); border-color: var(--red); }
-
-main { width: min(1120px, calc(100% - 48px)); margin: 0 auto; }
-
-.intro { padding: 72px 0 44px; border-bottom: 1px solid var(--ink); }
-.context, .section-index { margin: 0 0 8px; color: var(--red); font-size: 12px; font-weight: 700; }
-.context::before, .section-index::before { content: "# "; }
-
-h1, h2, h3, h4, p { text-wrap: pretty; }
-h1 { max-width: 16ch; margin: 0; font-size: 44px; line-height: 1.08; letter-spacing: -0.04em; }
-.lede { max-width: 68ch; margin: 22px 0 34px; color: var(--ink-soft); font-size: 16px; }
-
-.run-meta { margin: 0; display: flex; flex-wrap: wrap; gap: 20px 48px; }
-.run-meta p { margin: 0; display: grid; gap: 2px; }
-.run-meta span { color: var(--muted); font-size: 11px; }
-.run-meta strong { font-size: 12px; }
-
-.section { padding: 52px 0 56px; border-bottom: 1px solid var(--line); }
-.section-heading { display: flex; align-items: flex-end; justify-content: space-between; gap: 24px; margin-bottom: 24px; }
-.section-heading h2 { margin: 0; font-size: 23px; line-height: 1.2; letter-spacing: -0.035em; }
-.section-total, .result-count { margin: 0; color: var(--muted); font-size: 12px; }
-
-.coverage { display: inline-flex; align-items: center; gap: 8px; color: var(--ink-soft); font-size: 12px; }
-.coverage::before { content: ""; width: 8px; height: 8px; border-radius: 50%; background: var(--red); }
-
-.metrics { display: grid; grid-template-columns: repeat(4, 1fr); border-top: 1px solid var(--ink); border-bottom: 1px solid var(--line); }
-.metric { min-height: 132px; padding: 20px; border-right: 1px solid var(--line); display: flex; flex-direction: column; }
-.metric:last-child { border-right: 0; }
-.metric > span { color: var(--muted); font-size: 11px; }
-.metric strong { margin-top: auto; font-size: 25px; line-height: 1.2; letter-spacing: -0.04em; font-variant-numeric: tabular-nums; }
-.metric small { color: var(--muted); font-size: 10px; }
-.metric-unknown strong { color: var(--red-dark); font-size: 18px; }
-.source-note { margin: 12px 0 0; color: var(--muted); font-size: 11px; }
-.source-note code { color: var(--ink-soft); }
-
-.timing-list { padding: 0; margin: 0; list-style: none; border-top: 1px solid var(--ink); }
-.timing-list li { min-height: 74px; border-bottom: 1px solid var(--line); display: grid; grid-template-columns: 44px minmax(220px, 1fr) minmax(160px, 2fr); align-items: center; gap: 20px; }
-.timing-rank { color: var(--red); font-weight: 700; }
-.timing-list li > div { display: grid; }
-.timing-list li > div span { color: var(--muted); font-size: 11px; }
-.timing-bar {
-  width: 100%;
-  height: 7px;
-  border: 1px solid var(--line);
-  border-radius: 0;
-  background: var(--surface-soft);
-  appearance: none;
-}
-.timing-bar::-webkit-progress-bar { background: var(--surface-soft); }
-.timing-bar::-webkit-progress-value { background: var(--red); }
-.timing-bar::-moz-progress-bar { background: var(--red); }
-
-.filters { margin-bottom: 24px; padding: 14px 0; border-top: 1px solid var(--ink); border-bottom: 1px solid var(--line); display: flex; flex-wrap: wrap; gap: 16px 36px; }
+.site-header { min-height: 58px; padding: 0 32px; border-bottom: 1px solid var(--line); display: flex; align-items: center; justify-content: space-between; position: sticky; top: 0; z-index: 10; background: color-mix(in oklch, white 94%, transparent); backdrop-filter: blur(12px); }
+.brand { color: var(--ink); text-decoration: none; display: flex; align-items: center; gap: 10px; font-weight: 700; }
+.brand-mark { width: 25px; height: 25px; display: grid; place-items: center; background: var(--red); color: var(--ink); }
+.button { border: 1px solid var(--ink); padding: 7px 12px; background: var(--ink); color: white; cursor: pointer; }
+.button:hover { border-color: var(--red); background: var(--red); color: var(--ink); }
+
+main { width: min(1040px, calc(100% - 48px)); margin: auto; }
+.intro { padding: 58px 0 30px; border-bottom: 1px solid var(--ink); }
+.intro h1 { margin: 0; font-size: clamp(34px, 6vw, 48px); line-height: 1.05; letter-spacing: -0.045em; }
+.intro p { margin: 16px 0 0; color: var(--muted); font-size: 12px; }
+.section { padding: 42px 0; border-bottom: 1px solid var(--line); }
+.section-heading, .prompt-heading { display: flex; align-items: end; justify-content: space-between; gap: 20px; margin-bottom: 20px; }
+.section-heading h2, .prompt-heading h2 { margin: 0; font-size: 22px; letter-spacing: -0.03em; }
+.result-count, .section-total, .prompt-heading span, .prompt-actions span { color: var(--muted); font-size: 11px; }
+
+details > summary { cursor: pointer; color: var(--soft-ink); font-weight: 700; }
+.filter-panel { margin-bottom: 18px; border-block: 1px solid var(--line); padding: 11px 0; }
+.filters { display: flex; flex-wrap: wrap; gap: 12px 30px; padding-top: 12px; }
 .filter-group { display: flex; flex-wrap: wrap; align-items: center; gap: 6px; }
-.filter-group > span { margin-right: 5px; color: var(--muted); font-size: 11px; }
-.filter-group button {
-  border: 1px solid var(--line-strong);
-  border-radius: 999px;
-  padding: 4px 9px;
-  background: var(--surface);
-  color: var(--ink-soft);
-  font-size: 11px;
-  cursor: pointer;
-}
-.filter-group button:hover { border-color: var(--ink); color: var(--ink); }
-.filter-group button[aria-pressed="true"] { border-color: var(--ink); background: var(--ink); color: var(--surface); }
+.filter-group > span { margin-right: 4px; color: var(--muted); font-size: 11px; }
+.filter-group button { border: 1px solid var(--strong-line); border-radius: 99px; padding: 4px 8px; background: white; color: var(--soft-ink); font-size: 11px; cursor: pointer; }
+.filter-group button[aria-pressed="true"] { border-color: var(--ink); background: var(--ink); color: white; }
 
 .findings { border-top: 1px solid var(--ink); }
 .finding { border-bottom: 1px solid var(--ink); }
 .finding[hidden] { display: none; }
-.finding-header { padding: 26px 0 22px; display: grid; grid-template-columns: 62px minmax(0, 1fr) auto auto; gap: 20px; align-items: start; transition: padding 180ms ease-out; }
-.finding-id { color: var(--red); font-weight: 700; font-size: 12px; }
-.finding-title h3 { margin: 11px 0 0; font-size: 19px; line-height: 1.3; letter-spacing: -0.025em; }
-.tags { display: flex; flex-wrap: wrap; gap: 6px; }
-.tag { padding: 3px 7px; border: 1px solid var(--line-strong); font-size: 10px; line-height: 1.4; }
-.tag-question { border-color: var(--ink); background: var(--ink); color: var(--surface); }
-.tag-signal { border-color: var(--red); background: var(--red); color: var(--surface); }
-.tag-action { border-color: var(--ink); color: var(--ink); }
-
-.finding-score { margin: 0; display: flex; gap: 24px; }
-.finding-score p { margin: 0; display: grid; gap: 2px; }
-.finding-score span { color: var(--muted); font-size: 9px; }
-.finding-score strong { font-size: 11px; }
-
-.accept-toggle {
-  min-height: 31px;
-  padding: 5px 9px;
-  border: 1px solid var(--ink);
-  display: inline-flex;
-  align-items: center;
-  gap: 7px;
-  color: var(--ink);
-  font-size: 10px;
-  font-weight: 700;
-  cursor: pointer;
-}
-.accept-toggle:hover { border-color: var(--red); color: var(--red-dark); }
-.accept-toggle:focus-within { outline: 3px solid var(--focus); outline-offset: 3px; }
+.finding-header { padding: 22px 0; display: grid; grid-template-columns: 58px minmax(0, 1fr) auto; gap: 18px; align-items: start; }
+.finding-id { color: var(--red-dark); font-size: 11px; font-weight: 700; }
+.finding-title h3 { margin: 9px 0 0; font-size: 18px; line-height: 1.3; }
+.tags { display: flex; gap: 6px; }
+.tag { border: 1px solid var(--strong-line); padding: 3px 7px; font-size: 11px; }
+.tag-signal { border-color: var(--red); background: var(--red); color: var(--ink); }
+.accept-toggle { min-height: 31px; border: 1px solid var(--ink); padding: 5px 9px; display: flex; align-items: center; gap: 7px; font-size: 11px; font-weight: 700; cursor: pointer; }
 .accept-input { width: 14px; height: 14px; margin: 0; accent-color: var(--red); }
-.accept-input:focus-visible { outline: 0; }
-
-.finding-collapse { display: grid; grid-template-rows: 1fr; transition: grid-template-rows 200ms ease-out; }
-.finding-body { padding: 0 0 30px 82px; display: grid; grid-template-columns: minmax(240px, 0.8fr) minmax(360px, 1.45fr); gap: 36px; transition: padding 180ms ease-out; }
-.finding-id, .finding-title, .finding-score, .finding-collapse { transition: opacity 160ms ease-out; }
-.finding.is-accepted .finding-header { padding-block: 14px; }
-.finding.is-accepted .finding-collapse { grid-template-rows: 0fr; }
-.finding.is-accepted .finding-body { min-height: 0; padding-bottom: 0; overflow: hidden; }
-.finding.is-accepted .finding-id,
-.finding.is-accepted .finding-title,
-.finding.is-accepted .finding-score,
-.finding.is-accepted .finding-collapse { opacity: 0.48; }
+.finding-collapse { display: grid; grid-template-rows: 1fr; transition: grid-template-rows 180ms ease-out, opacity 180ms ease-out; }
+.finding-body { min-height: 0; padding: 0 0 26px 76px; overflow: hidden; }
+.finding.is-accepted .finding-header { padding-block: 12px; opacity: 0.55; }
+.finding.is-accepted .finding-collapse { grid-template-rows: 0fr; opacity: 0.35; }
 .finding.is-accepted .accept-toggle { border-color: var(--red); background: var(--red-soft); color: var(--red-dark); }
-.finding-copy h4 { margin: 0 0 5px; color: var(--muted); font-size: 10px; }
-.finding-copy p { max-width: 62ch; margin: 0 0 18px; color: var(--ink-soft); font-size: 12px; }
-.finding-copy .evidence { color: var(--ink); }
 
-.change { align-self: start; border: 1px solid var(--line-strong); background: var(--surface-soft); overflow: hidden; }
-.change-header { min-height: 38px; padding: 8px 11px; border-bottom: 1px solid var(--line); display: flex; justify-content: space-between; align-items: center; gap: 16px; font-size: 10px; }
+.change { border: 1px solid var(--strong-line); background: var(--soft-surface); overflow: hidden; }
+.change-header { min-height: 38px; padding: 8px 11px; border-bottom: 1px solid var(--line); display: flex; justify-content: space-between; gap: 16px; font-size: 11px; }
 .change-header span { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
-.copy-path { padding: 2px 0; border: 0; background: transparent; color: var(--red-dark); font-size: 9px; cursor: pointer; }
-.copy-path:hover { text-decoration: underline; }
-.diff { margin: 0; padding: 10px 0; overflow-x: auto; font: 11px/1.8 var(--mono); }
+.copy-path { border: 0; padding: 0; background: transparent; color: var(--red-dark); font-size: 11px; cursor: pointer; }
+.diff { margin: 0; padding: 9px 0; overflow-x: auto; font: 11px/1.8 var(--mono); }
 .diff code { display: block; min-width: max-content; }
 .diff-line { display: block; padding: 0 13px; }
 .diff-line b { display: inline-block; width: 18px; user-select: none; }
-.diff-del { background: var(--del-bg); color: var(--del-ink); }
-.diff-add { background: var(--add-bg); color: var(--add-ink); }
-.diff-del b { color: var(--red-dark); }
-.diff-add b { color: var(--green-dark); }
-.diff-context { color: var(--muted); }
-.empty-state { margin: 0; padding: 48px 0; border-bottom: 1px solid var(--ink); color: var(--muted); text-align: center; }
+.diff-del { background: var(--red-soft); color: var(--red-dark); }
+.diff-add { background: var(--green-soft); color: var(--ink); }
+.why { margin: 12px 0; color: var(--soft-ink); font-size: 11px; }
+.evidence-details { margin-top: 10px; font-size: 11px; }
+.evidence-details p { margin: 8px 0 0; padding: 10px; background: var(--soft-surface); }
+.empty-state { padding: 35px 0; color: var(--muted); text-align: center; }
+.limitations { margin-top: 18px; padding: 12px 0; border-bottom: 1px solid var(--line); font-size: 11px; }
 
 .table-wrap { overflow-x: auto; border-top: 1px solid var(--ink); }
-table { width: 100%; min-width: 680px; border-collapse: collapse; font-size: 12px; text-align: left; }
-th { padding: 12px 14px; border-bottom: 1px solid var(--ink); color: var(--muted); font-size: 10px; font-weight: 500; }
-td { padding: 16px 14px; border-bottom: 1px solid var(--line); vertical-align: top; }
+table { width: 100%; min-width: 650px; border-collapse: collapse; font-size: 11px; text-align: left; }
+th { padding: 10px 12px; border-bottom: 1px solid var(--ink); color: var(--muted); font-size: 11px; font-weight: 500; }
+td { padding: 13px 12px; border-bottom: 1px solid var(--line); vertical-align: top; }
 th:first-child, td:first-child { padding-left: 0; }
 th:last-child, td:last-child { padding-right: 0; }
 td code { color: var(--red-dark); font-weight: 700; }
 
-.prompt-builder { margin: 64px 0; padding: 28px 0 0; border-top: 4px solid var(--red); }
-.prompt-heading { display: flex; align-items: end; justify-content: space-between; gap: 24px; }
-.prompt-heading p { margin: 0 0 5px; color: var(--red); font-size: 11px; font-weight: 700; }
-.prompt-heading h2 { margin: 0; font-size: 22px; line-height: 1.3; letter-spacing: -0.025em; }
-.prompt-heading > span, .prompt-actions > span { color: var(--muted); font-size: 11px; }
-.prompt-help { max-width: 72ch; margin: 18px 0 12px; color: var(--ink-soft); font-size: 12px; }
-#execution-prompt {
-  width: 100%;
-  min-height: 230px;
-  padding: 16px;
-  border: 1px solid var(--ink);
-  border-radius: 0;
-  resize: vertical;
-  background: var(--surface-soft);
-  color: var(--ink);
-  font: 12px/1.7 var(--mono);
-}
-#execution-prompt:focus { outline: 3px solid var(--focus); outline-offset: 3px; }
+.raw-section { padding: 20px 0; }
+.raw-section > summary { font-size: 15px; }
+.raw-section[open] > summary { margin-bottom: 18px; }
+.metrics { display: grid; grid-template-columns: repeat(4, 1fr); border-top: 1px solid var(--ink); }
+.metrics p { margin: 0; padding: 14px; border-right: 1px solid var(--line); display: grid; gap: 5px; }
+.metrics p:last-child { border-right: 0; }
+.metrics span { color: var(--muted); font-size: 11px; }
+.metrics strong { font-size: 18px; }
+.raw-section h3 { margin: 20px 0 5px; font-size: 12px; }
+.raw-section p { color: var(--soft-ink); font-size: 11px; }
+.source-records { margin-top: 14px; font-size: 11px; }
+
+.prompt-builder { margin: 54px 0; padding-top: 24px; border-top: 4px solid var(--red); }
+#execution-prompt { width: 100%; min-height: 220px; padding: 15px; border: 1px solid var(--ink); resize: vertical; background: var(--soft-surface); color: var(--ink); font: 12px/1.7 var(--mono); }
 .prompt-actions { margin-top: 10px; display: flex; align-items: center; justify-content: space-between; gap: 20px; }
 
-@media (max-width: 820px) {
-  .metrics { grid-template-columns: repeat(2, 1fr); }
-  .metric:nth-child(2) { border-right: 0; }
-  .metric:nth-child(-n + 2) { border-bottom: 1px solid var(--line); }
-  .finding-header { grid-template-columns: 50px minmax(0, 1fr) auto; }
-  .finding-score { grid-column: 2; }
-  .accept-toggle { grid-column: 3; grid-row: 1 / span 2; }
-  .finding-body { padding-left: 70px; grid-template-columns: 1fr; }
-}
-
-@media (max-width: 580px) {
+@media (max-width: 680px) {
   .site-header { padding: 0 16px; }
-  .run-id { display: none; }
-  main { width: min(100% - 32px, 1120px); }
-  .intro { padding-top: 48px; }
-  h1 { font-size: 36px; }
-  .section-heading { align-items: start; flex-direction: column; gap: 8px; }
-  .metrics { grid-template-columns: 1fr; }
-  .metric { min-height: 105px; border-right: 0; border-bottom: 1px solid var(--line); }
-  .metric:last-child { border-bottom: 0; }
-  .timing-list li { padding: 14px 0; grid-template-columns: 30px 1fr; gap: 8px 12px; }
-  .timing-bar { grid-column: 2; }
+  main { width: min(100% - 32px, 1040px); }
   .finding-header { grid-template-columns: 1fr; gap: 8px; }
-  .finding-score { grid-column: 1; flex-wrap: wrap; }
-  .accept-toggle { grid-column: 1; grid-row: auto; justify-self: start; }
+  .accept-toggle { justify-self: start; }
   .finding-body { padding-left: 0; }
-  .finding-id { order: -1; }
-  .prompt-heading { align-items: start; flex-direction: column; gap: 6px; }
-  .prompt-actions { align-items: stretch; flex-direction: column; }
-  .prompt-actions .button { align-self: flex-start; }
-}
-
-@media (prefers-reduced-motion: reduce) {
-  html { scroll-behavior: auto; }
-  .finding-header, .finding-id, .finding-title, .finding-score, .finding-collapse { transition-duration: 0.001ms; }
+  .metrics { grid-template-columns: repeat(2, 1fr); }
+  .metrics p:nth-child(2) { border-right: 0; }
+  .section-heading, .prompt-heading, .prompt-actions { align-items: start; flex-direction: column; }
 }
 
+@media (prefers-reduced-motion: reduce) { .finding-collapse { transition-duration: 0.001ms; } }
 @media print {
-  .site-header, .filters, .copy-path, .accept-toggle, .prompt-actions { display: none !important; }
-  body { color: black; font-size: 11px; }
+  .site-header, .filter-panel, .copy-path, .accept-toggle, .prompt-actions { display: none !important; }
   main { width: 100%; }
-  .intro { padding-top: 0; }
-  .section, .finding { break-inside: avoid; }
   .finding[hidden] { display: block; }
   .finding.is-accepted { display: none; }
-  .change { background: white; }
 }
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report.js b/plugins/aidd-refine/skills/05-improve/assets/report.js
index 72770d06f..f36fda79e 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report.js
+++ b/plugins/aidd-refine/skills/05-improve/assets/report.js
@@ -1,12 +1,36 @@
 (() => {
+  const promptAnchor = "Changements acceptés :";
+
+  function composePrompt(value, accepted, savedLines = new Map(), knownIds = new Set(accepted.map(({ id }) => id))) {
+    const generatedPattern = /^- \[([^\]]+)\] /;
+    const original = value.split("\n");
+
+    for (const line of original) {
+      const match = line.match(generatedPattern);
+      if (match && knownIds.has(match[1])) savedLines.set(match[1], line);
+    }
+
+    const lines = original.filter((line) => {
+      const match = line.match(generatedPattern);
+      return !match || !knownIds.has(match[1]);
+    });
+    const generated = accepted.map(({ id, prompt }) => savedLines.get(id) || `- [${id}] ${prompt}`);
+    const anchorIndex = lines.findIndex((line) => line.trim() === promptAnchor);
+    lines.splice(anchorIndex >= 0 ? anchorIndex + 1 : lines.length, 0, ...generated);
+    return lines.join("\n");
+  }
+
+  if (typeof module !== "undefined") module.exports = { composePrompt };
+  if (typeof document === "undefined") return;
+
   const state = { signal: "all", target: "all" };
   const findings = [...document.querySelectorAll(".finding")];
   const count = document.querySelector("#result-count");
   const empty = document.querySelector("#empty-state");
   const prompt = document.querySelector("#execution-prompt");
   const acceptedCount = document.querySelector("#accepted-count");
-  const promptAnchor = "Changements acceptés :";
-  const closingPrompt = "Une fois le travail validé";
+  const savedPromptLines = new Map();
+  const findingIds = new Set(findings.map((finding) => finding.querySelector(".finding-id").textContent.trim()));
 
   function render() {
     let visible = 0;
@@ -44,22 +68,13 @@
     });
   });
 
-  function updatePrompt(changedFinding) {
+  function updatePrompt() {
     const accepted = findings.filter((finding) => finding.querySelector(".accept-input").checked);
-    const changedInput = changedFinding.querySelector(".accept-input");
-    const id = changedFinding.querySelector(".finding-id").textContent.trim();
-    const prefix = `- [${id}] `;
-    const generatedLine = `${prefix}${changedInput.dataset.prompt}`;
-    const lines = prompt.value.split("\n").filter((line) => !line.startsWith(prefix));
-    const anchorIndex = lines.findIndex((line) => line.trim() === promptAnchor);
-
-    if (changedInput.checked) {
-      const closingIndex = lines.findIndex((line) => line.startsWith(closingPrompt));
-      let insertAt = closingIndex >= 0 ? closingIndex : anchorIndex + 1;
-      if (lines[insertAt - 1] === "") insertAt -= 1;
-      lines.splice(Math.max(insertAt, 0), 0, generatedLine);
-    }
-    prompt.value = lines.join("\n");
+    const entries = accepted.map((finding) => ({
+      id: finding.querySelector(".finding-id").textContent.trim(),
+      prompt: finding.querySelector(".accept-input").dataset.prompt,
+    }));
+    prompt.value = composePrompt(prompt.value, entries, savedPromptLines, findingIds);
 
     const total = accepted.length;
     acceptedCount.textContent = `${total} recommandation${total === 1 ? "" : "s"} acceptée${total === 1 ? "" : "s"}`;
@@ -70,7 +85,7 @@
       const finding = input.closest(".finding");
       finding.classList.toggle("is-accepted", input.checked);
       finding.querySelector(".accept-toggle span").textContent = input.checked ? "Accepté" : "Accepter";
-      updatePrompt(finding);
+      updatePrompt();
     });
   });
 
diff --git a/scripts/__tests__/improve-report.test.js b/scripts/__tests__/improve-report.test.js
new file mode 100644
index 000000000..374699090
--- /dev/null
+++ b/scripts/__tests__/improve-report.test.js
@@ -0,0 +1,46 @@
+const assert = require('node:assert/strict');
+const { resolve } = require('node:path');
+const test = require('node:test');
+
+const { composePrompt } = require(resolve(
+  __dirname,
+  '../../plugins/aidd-refine/skills/05-improve/assets/report.js',
+));
+
+test('accepted edits keep report order and preserve manual prompt text', () => {
+  const saved = new Map();
+  const known = new Set(['F-001', 'F-002']);
+  const base = [
+    'User introduction.',
+    '',
+    'Changements acceptés :',
+    '- [NOTE] Keep this manual note.',
+    '',
+    'User closing.',
+  ].join('\n');
+
+  let value = composePrompt(base, [{ id: 'F-002', prompt: 'Second edit.' }], saved, known);
+  value = value.replace('- [F-002] Second edit.', '- [F-002] Second edit, manually refined.');
+  value = composePrompt(value, [
+    { id: 'F-001', prompt: 'First edit.' },
+    { id: 'F-002', prompt: 'Second edit.' },
+  ], saved, known);
+
+  assert.ok(value.indexOf('[F-001]') < value.indexOf('[F-002]'));
+  assert.match(value, /\[F-002\] Second edit, manually refined\./);
+  assert.match(value, /^User introduction\./);
+  assert.match(value, /- \[NOTE\] Keep this manual note\./);
+  assert.match(value, /User closing\.$/);
+
+  value = composePrompt(value, [{ id: 'F-001', prompt: 'First edit.' }], saved, known);
+  assert.doesNotMatch(value, /\[F-002\]/);
+  value = composePrompt(value, [
+    { id: 'F-001', prompt: 'First edit.' },
+    { id: 'F-002', prompt: 'Second edit.' },
+  ], saved, known);
+  assert.match(value, /\[F-002\] Second edit, manually refined\./);
+  assert.equal(composePrompt(value, [
+    { id: 'F-001', prompt: 'First edit.' },
+    { id: 'F-002', prompt: 'Second edit.' },
+  ], saved, known), value);
+});

From 521bf54b423be902f8054cce8259f93ee69de3ed Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Mon, 5 Oct 2026 10:25:44 +0700
Subject: [PATCH 5/7] fix(test): isolate hook probes from source scanners

---
 scripts/__tests__/check-written-file-hook.test.js | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/scripts/__tests__/check-written-file-hook.test.js b/scripts/__tests__/check-written-file-hook.test.js
index 16b8a9242..260f50cd7 100644
--- a/scripts/__tests__/check-written-file-hook.test.js
+++ b/scripts/__tests__/check-written-file-hook.test.js
@@ -14,7 +14,8 @@ function runHook(payload) {
 }
 
 function probe(name, content) {
-  const file = path.join(root, "cli/src", `.hook-probe-${process.pid}-${name}.ts`);
+  // Keep transient probes outside directories scanned by repository-wide tests.
+  const file = path.join(root, "cli", `.hook-probe-${process.pid}-${name}.ts`);
   fs.writeFileSync(file, content);
   return file;
 }

From 9a9cffdcd4b9cb0feb932794001ea83e6f533233 Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Mon, 5 Oct 2026 10:27:02 +0700
Subject: [PATCH 6/7] refactor(aidd-refine): scope improve actions and report

---
 .../aidd-refine/skills/05-improve/SKILL.md    | 17 ++--
 .../actions/01-read-conversation.md           | 23 ++---
 .../skills/05-improve/actions/02-recommend.md | 40 +++------
 .../05-improve/actions/03-target-edits.md     | 23 ++---
 .../05-improve/assets/conversation-sources.md |  8 --
 .../05-improve/assets/report-template.html    | 86 +++++++++----------
 .../skills/05-improve/assets/report.js        | 29 ++++---
 scripts/__tests__/improve-report.test.js      | 35 +++++++-
 8 files changed, 120 insertions(+), 141 deletions(-)

diff --git a/plugins/aidd-refine/skills/05-improve/SKILL.md b/plugins/aidd-refine/skills/05-improve/SKILL.md
index 509d26ec2..fd24b0e7a 100644
--- a/plugins/aidd-refine/skills/05-improve/SKILL.md
+++ b/plugins/aidd-refine/skills/05-improve/SKILL.md
@@ -15,17 +15,10 @@ flowchart LR
 
 ## Actions
 
-Run all three actions without confirmation. Read only the next action file.
+Run all three actions in order without confirmation. Read only the next file in `actions/`. Keep the project read-only; write only the unique temporary report.
 
-| Action | Does |
+| Order | Action |
 | --- | --- |
-| read-conversation | freeze complete evidence and measure visible cost |
-| recommend | analyze relevant scopes and merge grounded findings |
-| target-edits | render minimal edits and an executable prompt |
-
-## Transversal rules
-
-- Stop if the exact complete transcript is unavailable.
-- Assess conversation behavior, invoked skill resources, applicable project instructions, and task-relevant indexed memory.
-- Never invent metrics, coverage, status, or private reasoning.
-- Keep the project read-only; write only the unique temporary report.
+| 1 | `01-read-conversation.md` |
+| 2 | `02-recommend.md` |
+| 3 | `03-target-edits.md` |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
index d23d1d1e6..ede26347b 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
@@ -8,25 +8,16 @@ A current conversation, an exact conversation ID, or a complete transcript expor
 
 ## Output
 
-An evidence boundary, a `## Timing` table with `Activity | Observed time | Share | Evidence`, a `## Usage` table with `Metric | Value | Evidence`, and a `Scope | Status | Evidence` index.
+An evidence boundary, a `## Timing` table with `Activity | Observed time | Share | Evidence`, a `## Usage` table with `Metric | Value | Evidence`, and a scope index of encountered resources.
 
 ## Process
 
-1. **Resolve.** Use the host route in [conversation sources](../assets/conversation-sources.md); stop unless it yields the exact complete transcript.
+1. **Resolve.** Use the host route in [conversation sources](../assets/conversation-sources.md). Prefer its complete export or host reader. Match the exact session ID before direct storage; search only for that ID, never unrelated sessions, configuration, or authentication. Stop unless the exact complete transcript is available.
 2. **Freeze.** End before this invocation, or at the export boundary. Exclude every `improve` invocation and report.
-3. **Read once.** Load all in-scope messages, tool calls, results, and timestamps. Index relevant turns, invoked skills with used actions or resources, and applicable project instructions.
-4. **Context.** Read indexed skill resources and applicable `AGENTS.md`. Use the project memory index to read only task-relevant memory; never scan the whole memory library. Mark only read files `checked`; mark unresolved or skipped scopes `missing` or `not reviewed`.
-5. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
+3. **Read once.** Load every in-scope message, tool call, result, and timestamp. Index relevant turns; invoked skills and their used actions or resources; applicable `AGENTS.md`; and task-relevant memory references. Do not read maintained sources yet.
+4. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
    - Group tool time as `research and diagnosis`, `implementation`, `validation`, or `unattributed`.
-   - Include tokens, requests, and cost only when exposed.
+   - Include tokens, requests, and cost only when exposed; otherwise mark them `unavailable`.
    - Separate background-process lifetime from blocking time.
-6. **Render.** Order known activity times descending; mark missing metrics `unavailable`. Never call unattributed or unavailable time private reasoning.
-
-## Test
-
-| Case | Pass |
-| --- | --- |
-| Conversation | exact complete transcript; frozen boundary; no `improve` evidence |
-| Metric | cited host record, timestamp, or elapsed record; otherwise `unavailable` |
-| Time | visible categories only; background lifetime separated; no private-reasoning claim |
-| Scope | exact turns or paths; only read files are `checked` |
+   - Treat private reasoning duration as unavailable unless exposed directly; never infer it.
+5. **Deliver.** Order known activity times descending and cite every value.
diff --git a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
index 10a5e5319..9180a14d0 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
@@ -4,50 +4,34 @@ Reduce avoidable work without reducing reliability.
 
 ## Input
 
-The evidence boundary, timing and usage tables, complete conversation, and scope index.
+The evidence boundary, timing and usage tables, complete conversation, and resource index.
 
 ## Output
 
-A `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Source | Saving`, plus non-actionable limitations.
+An updated `Scope | Status | Evidence` index, a `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Source | Saving`, and non-actionable limitations.
 
 ## Process
 
-1. **Scope.** Always assess `behavior`; add `skill` or `knowledge` for indexed skill resources, instructions, or memory.
-2. **Dispatch.** With multiple scopes and isolated or minimal context, run one read-only analyst per scope in parallel. Otherwise analyze locally.
+1. **Verify.** Resolve and read indexed maintained skill resources, applicable `AGENTS.md`, and task-relevant memory references once; reuse a still-valid read. Add and read only relevant paths from the project memory index; never scan unrelated memory. Neither judge from nor target caches or installs. Account for host transforms. Mark only read sources `checked`; mark unresolved or skipped ones `missing` or `not reviewed`.
+2. **Scope.** Always assess `behavior`; add `skill` or `knowledge` only for checked indexed sources.
+3. **Dispatch.** With multiple isolated scopes, run one read-only analyst per scope in parallel. Otherwise analyze locally.
    - Prefer a cheap model and low reasoning effort when per-agent overrides are supported; otherwise inherit host defaults.
-   - Give the behavior analyst the frozen transcript; give each artifact analyst the boundary, indexed turns, and paths.
-   - Give every analyst steps 3–7; require table rows and no writes.
-3. **Reflect.** Ask yourself, for each scope; keep this assessment internal:
+   - Give the behavior analyst the frozen transcript. Give artifact analysts the boundary, indexed turns, and verified excerpts with evidence; allow a targeted read-only refresh only when validity is uncertain.
+   - Give every analyst steps 4–6 and require table rows.
+4. **Reflect.** Ask yourself, for each scope; keep this assessment internal:
    - How could the next run be faster or better?
    - What should be removed or clarified? What was counterproductive?
    - Where should change live? How could it save time or tokens?
    - Which back-and-forth, bottlenecks, or tool calls could be removed, batched, parallelized, or replaced?
-4. **Verify.** Resolve and read each maintained source before judging it; never target caches or installs. Account for intentional host transforms. Judge only `checked` sources; an installed artifact never makes its source `checked`.
 5. **Assess.** Use `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
    - Compare observed work with the shortest reliable path to the requested result. Keep checks for mutable state or unresolved uncertainty; reuse still-valid results instead of repeating research.
    - Raw call count never proves waste. Ground the smallest general correction in evidence.
    - Recommend memory only for concise, durable knowledge that prevents recurring rediscovery, never transient state or one-off implementation details.
    - For skills, especially knowledge: delete evidence-backed waste or inconsistency first, without quota; then consolidate or clarify.
-   - Reuse or strengthen existing content; add only for a demonstrated gap it cannot cover. Preserve useful context and requirements; use `correct` when it already suffices.
+   - Reuse or strengthen existing content; add only for a demonstrated gap it cannot cover. Preserve useful context and requirements; classify sufficient content as `correct` => `no change`.
    - Claim time savings from blocking-time or resource-impact evidence, never background-process uptime alone.
-   - `correct` means `no change`, never a recommendation.
-6. **Merge.** Deduplicate, then verify cited evidence against the frozen source.
-7. **Render.** Order by focus, then `behavior`, `skill`, `knowledge`. Report findings, not the questionnaire. Use the fewest actionable words.
-   - Every recommendation, including behavior or tooling, needs an exact maintained source path, exact current excerpt, minimal replacement or explicit `+`/`−` edit, and a short target/action prompt.
-   - If no source resolves or no useful persistent edit exists, render a collapsed limitation, not an accept-toggle recommendation.
+6. **Deliver.** Deduplicate, verify cited evidence, and order by focus, then `behavior`, `skill`, `knowledge`. Report findings, not the questionnaire, in the fewest actionable words.
+   - Every recommendation needs an exact maintained source path, current excerpt, minimal correction or explicit `+`/`−` edit, and a short target/action prompt.
+   - If no source resolves or no useful persistent edit exists, report a non-actionable limitation.
    - Type: `skill`, `behavior`, `knowledge`, or `tooling`.
    - Saving: `time`, `tokens`, `both`, or `unknown`.
-   - Use `no change` when evidence supports none.
-
-## Test
-
-| Case | Pass |
-| --- | --- |
-| Dispatch | parallel only for multiple isolated scopes; otherwise local |
-| Analysis | internal assessment answers every question with exact evidence or `no change` |
-| Efficiency | shortest reliable path; necessary checks retained; repeated work evidenced |
-| Finding | exact evidence, maintained source path, minimal edit, and target/action prompt |
-| No finding | `no change`; no false recommendation |
-| Coverage | verdicts only for `checked` files; other scopes stay explicit |
-| Limitation | unresolved or non-persistent work is not actionable |
-| Output | allowed labels, types, savings, and stable order; no `improve` evidence |
diff --git a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
index ddc8cc87a..ec8f9c812 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
@@ -12,22 +12,9 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 
 ## Process
 
-1. **Target.** Map every recommendation to its maintained source path. Build `Fichier | + Ajout | − Retrait ou clarification`, consolidate targets, and show `Aucun fichier recommandé | — | —` when empty. Recommend only; never apply.
-2. **Render.** Fill [the report template](../assets/report-template.html) with measured values and grounded findings; copy its local CSS and JavaScript.
-   - Remove samples. Escape every injected value; allow only template markup and local assets.
-   - Put findings first. Preserve diagnostic/type tags, collapsible filters, accept controls, target table, and prompt.
-   - Show each source path and minimal edit. Keep raw evidence, secondary metrics, timing, coverage, source records, and limitations collapsed.
-   - Remove repeated metadata, focus/action tags, score labels, verbose help, and redundant run metadata; put the run ID in collapsed source records.
-   - Keep `data-prompt` and editable execution instructions as short as possible without losing targets or actions.
-   - Write only to the allowed unique temporary directory.
+1. **Target.** Consolidate the provided recommendations by source into `File | + Added | − Removed or clarified`; use `No recommended file | — | —` when empty. Never resolve, judge, or apply sources here.
+2. **Render.** Fill [the report template](../assets/report-template.html) from the provided data; remove its sample and copy its local CSS and JavaScript into the allowed unique temporary directory.
+   - Escape injected values; allow only template markup and local assets. Localize the HTML `lang`, visible text, and dynamic `data-label-*` text to the user's language.
+   - Put findings first with diagnostic/type tags, filters, accept controls, the target table, and an editable global prompt. Copy each recommendation's short target/action prompt into `data-prompt` without redrafting it.
+   - Show source paths and minimal edits. Collapse raw evidence, secondary metrics, timing, coverage, source records, and limitations; keep the run ID in source records and avoid repeated metadata.
 3. **Return.** Return its path. Ask the user, in their language, which small general change in intent to apply next run for cumulative, measurable improvement.
-
-## Test
-
-| Case | Pass |
-| --- | --- |
-| Edit table | every actionable finding has a maintained source row; consolidated or empty |
-| Report | no samples; unavailable values and scope coverage named; injected evidence inert |
-| Assets | HTML, CSS, and JavaScript resolve locally only |
-| Findings | exact source edit and accept control; evidence and limitations collapse |
-| Close | path returned; next intent requested in the user's language |
diff --git a/plugins/aidd-refine/skills/05-improve/assets/conversation-sources.md b/plugins/aidd-refine/skills/05-improve/assets/conversation-sources.md
index 329cd7da7..c4a689fbe 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/conversation-sources.md
+++ b/plugins/aidd-refine/skills/05-improve/assets/conversation-sources.md
@@ -7,11 +7,3 @@ Use this static routing map to load one complete conversation. Do not fill or co
 | Codex | host thread reader with the exact thread ID, then TUI `/export` Markdown | `CODEX_HOME/history.jsonl`, normally `~/.codex/history.jsonl` | TUI `/status` estimated thread credits or cost; tool elapsed records when exposed |
 | Claude Code | `/export` text, or the documented script interface for an exact session ID | `~/.claude/projects/{project}/{session-id}.jsonl` | transcript timestamps and tool records when exposed |
 | OpenCode | `opencode export {session-id} --sanitize` JSON | `~/.local/share/opencode/project/{project-slug}/storage/` | `opencode stats`; `~/.local/share/opencode/log/` |
-
-## Rules
-
-- Prefer the complete export or host reader over direct storage parsing.
-- Use direct storage only after matching one exact session ID.
-- Search shared history or logs only for the exact session ID and ignore unrelated records.
-- Never enumerate unrelated sessions or inspect configuration or authentication files.
-- Treat private reasoning duration as unavailable unless the host exposes it directly.
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report-template.html b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
index 0d0f26cae..bc239f99e 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report-template.html
+++ b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
@@ -1,11 +1,5 @@
 <!doctype html>
-<!--
-  Replace all samples with current evidence. Each actionable `.finding` needs
-  diagnostic/type tags, a maintained source path, minimal edit, evidence, and
-  target/action data-prompt. Put unresolved or non-persistent items in collapsed
-  limitations. Keep report.css/report.js local. Remove this comment in output.
--->
-<html lang="fr">
+<html lang="en">
   <head>
     <meta charset="utf-8" />
     <meta name="viewport" content="width=device-width, initial-scale=1" />
@@ -17,33 +11,33 @@
   <body>
     <header class="site-header">
       <a class="brand" href="#report" aria-label="AIDD Improve report"><span class="brand-mark" aria-hidden="true">A</span><span>AIDD / improve</span></a>
-      <div class="header-actions"><button class="button" id="print-report" type="button">Imprimer</button></div>
+      <div class="header-actions"><button class="button" id="print-report" type="button">Print</button></div>
     </header>
 
-    <main id="report">
+    <main id="report" data-label-result-one="{count} recommendation" data-label-result-other="{count} recommendations" data-label-accepted-one="{count} recommendation accepted" data-label-accepted-other="{count} recommendations accepted">
       <section class="intro" aria-labelledby="report-title">
-        <h1 id="report-title">Rapport d’amélioration</h1>
-        <p>Exemple · conversation actuelle · couverture partielle · 1 recommandation</p>
+        <h1 id="report-title">Improve report</h1>
+        <p>Example only · fictional content · partial coverage · 1 recommendation</p>
       </section>
 
       <section class="section findings-section" aria-labelledby="findings-title">
-        <div class="section-heading"><h2 id="findings-title">Changements recommandés</h2><p id="result-count" class="result-count" aria-live="polite">1 recommandation</p></div>
+        <div class="section-heading"><h2 id="findings-title">Recommended changes</h2><p id="result-count" class="result-count" aria-live="polite">1 recommendation</p></div>
 
         <details class="filter-panel">
-          <summary>Filtres</summary>
-          <div class="filters" aria-label="Filtrer les recommandations">
+          <summary>Filters</summary>
+          <div class="filters" aria-label="Filter recommendations">
             <div class="filter-group" role="group" aria-label="Diagnostic">
               <span>Diagnostic</span>
-              <button type="button" data-filter="signal" data-value="all" aria-pressed="true">Tous</button>
-              <button type="button" data-filter="signal" data-value="obsolete" aria-pressed="false">Obsolète</button>
-              <button type="button" data-filter="signal" data-value="over-specific-or-time-bound" aria-pressed="false">Trop spécifique</button>
-              <button type="button" data-filter="signal" data-value="duplicate" aria-pressed="false">Doublon</button>
-              <button type="button" data-filter="signal" data-value="inconsistent" aria-pressed="false">Incohérent</button>
-              <button type="button" data-filter="signal" data-value="counterproductive" aria-pressed="false">Contre-productif</button>
+              <button type="button" data-filter="signal" data-value="all" aria-pressed="true">All</button>
+              <button type="button" data-filter="signal" data-value="obsolete" aria-pressed="false">Obsolete</button>
+              <button type="button" data-filter="signal" data-value="over-specific-or-time-bound" aria-pressed="false">Over-specific</button>
+              <button type="button" data-filter="signal" data-value="duplicate" aria-pressed="false">Duplicate</button>
+              <button type="button" data-filter="signal" data-value="inconsistent" aria-pressed="false">Inconsistent</button>
+              <button type="button" data-filter="signal" data-value="counterproductive" aria-pressed="false">Counterproductive</button>
             </div>
-            <div class="filter-group" role="group" aria-label="Type de cible">
-              <span>Cible</span>
-              <button type="button" data-filter="target" data-value="all" aria-pressed="true">Toutes</button>
+            <div class="filter-group" role="group" aria-label="Target type">
+              <span>Target</span>
+              <button type="button" data-filter="target" data-value="all" aria-pressed="true">All</button>
               <button type="button" data-filter="target" data-value="behavior" aria-pressed="false">Behavior</button>
               <button type="button" data-filter="target" data-value="skill" aria-pressed="false">Skill</button>
               <button type="button" data-filter="target" data-value="knowledge" aria-pressed="false">Knowledge</button>
@@ -53,52 +47,52 @@ <h1 id="report-title">Rapport d’amélioration</h1>
         </details>
 
         <div class="findings" id="findings-list">
-          <article class="finding" data-signal="inconsistent" data-target="skill">
+          <article class="finding" data-signal="duplicate" data-target="knowledge">
             <header class="finding-header">
               <div class="finding-id">F-001</div>
-              <div class="finding-title"><div class="tags"><span class="tag tag-signal">Incohérent</span><span class="tag">Skill</span></div><h3>Exemple · préciser la preuve d’économie</h3></div>
-              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="Dans plugins/aidd-refine/skills/05-improve/actions/02-recommend.md, préciser que l’uptime seul ne prouve aucune économie."><span>Accepter</span></label>
+              <div class="finding-title"><div class="tags"><span class="tag tag-signal">Duplicate</span><span class="tag">Knowledge</span></div><h3>Example · remove a fictional duplicate rule</h3></div>
+              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="In docs/example-guidance.md, remove the second duplicate 'Validate before publishing.' rule." data-label-off="Accept" data-label-on="Accepted"><span>Accept</span></label>
             </header>
             <div class="finding-collapse"><div class="finding-body">
               <div class="change">
-                <div class="change-header"><span>plugins/aidd-refine/skills/05-improve/actions/02-recommend.md</span><button class="copy-path" type="button" data-copy="plugins/aidd-refine/skills/05-improve/actions/02-recommend.md">Copier</button></div>
-                <pre class="diff" aria-label="Modification proposée pour 02-recommend.md"><code><span class="diff-line diff-del"><b>−</b>Claim time savings from blocking-time or resource-impact evidence, never background-process uptime alone.</span>
-<span class="diff-line diff-add"><b>+</b>Claim time savings only from blocking-time or resource-impact evidence; background uptime alone proves none.</span></code></pre>
+                <div class="change-header"><span>docs/example-guidance.md</span><button class="copy-path" type="button" data-copy="docs/example-guidance.md" data-label-success="Copied" data-label-error="Copy unavailable">Copy</button></div>
+                <pre class="diff" aria-label="Proposed change for example-guidance.md"><code><span class="diff-line"><b> </b>Validate before publishing.</span>
+<span class="diff-line diff-del"><b>−</b>Validate before publishing.</span></code></pre>
               </div>
-              <p class="why">Sépare durée observée et impact réel.</p>
-              <details class="evidence-details"><summary>Preuve</summary><p>Le processus observé restait actif sans bloquer la suite.</p></details>
+              <p class="why">One rule carries the same requirement.</p>
+              <details class="evidence-details"><summary>Evidence</summary><p>The fictional source contains the same rule twice.</p></details>
             </div></div>
           </article>
 
         </div>
-        <p class="empty-state" id="empty-state" hidden>Aucune recommandation ne correspond.</p>
+        <p class="empty-state" id="empty-state" hidden>No recommendations match.</p>
 
-        <details class="limitations"><summary>Limites non actionnables · 1</summary><p>Une source citée n’a pas pu être résolue ; aucun changement n’est proposé.</p></details>
+        <details class="limitations"><summary>Non-actionable limitations · 1</summary><p>A cited source could not be resolved, so no change is proposed.</p></details>
       </section>
 
       <section class="section target-section" aria-labelledby="target-title">
-        <div class="section-heading"><h2 id="target-title">Fichiers ciblés</h2><span class="section-total">1 fichier</span></div>
-        <div class="table-wrap"><table><thead><tr><th>Fichier</th><th>+ Ajout</th><th>− Retrait ou clarification</th></tr></thead><tbody>
-          <tr><td><code>plugins/aidd-refine/skills/05-improve/actions/02-recommend.md</code></td><td>« uptime seul = aucune économie prouvée »</td><td>Formulation ambiguë</td></tr>
+        <div class="section-heading"><h2 id="target-title">Target files</h2><span class="section-total">1 file</span></div>
+        <div class="table-wrap"><table><thead><tr><th>File</th><th>+ Added</th><th>− Removed or clarified</th></tr></thead><tbody>
+          <tr><td><code>docs/example-guidance.md</code></td><td>—</td><td>Second duplicate rule</td></tr>
         </tbody></table></div>
       </section>
 
       <details class="section raw-section">
-        <summary>Mesures, chronologie et couverture</summary>
-        <div class="metrics"><p><span>Durée active</span><strong>24 min</strong></p><p><span>Tokens</span><strong>128,4 k</strong></p><p><span>Requêtes</span><strong>42</strong></p><p><span>Coût</span><strong>Inconnu</strong></p></div>
-        <div class="table-wrap"><table><thead><tr><th>Activité</th><th>Temps</th><th>Part</th></tr></thead><tbody><tr><td>Recherche et diagnostic</td><td>12 min</td><td>50 %</td></tr><tr><td>Implémentation</td><td>8 min</td><td>33 %</td></tr><tr><td>Validation</td><td>4 min</td><td>17 %</td></tr></tbody></table></div>
-        <h3>Couverture</h3><p><code>AGENTS.md</code> · checked<br /><code>aidd_docs/memory/testing.md</code> · not reviewed<br /><code>skill export</code> · missing</p>
-        <details class="source-records"><summary>Sources brutes</summary><p><code>run_2026-09-12_01</code> · <code>aidd telemetry report v8</code> · 2026-09-12T02:00Z → 03:00Z</p></details>
+        <summary>Metrics, timeline, and coverage</summary>
+        <div class="metrics"><p><span>Active time</span><strong>24 min</strong></p><p><span>Tokens</span><strong>128.4k</strong></p><p><span>Requests</span><strong>42</strong></p><p><span>Cost</span><strong>Unknown</strong></p></div>
+        <div class="table-wrap"><table><thead><tr><th>Activity</th><th>Time</th><th>Share</th></tr></thead><tbody><tr><td>Research and diagnosis</td><td>12 min</td><td>50%</td></tr><tr><td>Implementation</td><td>8 min</td><td>33%</td></tr><tr><td>Validation</td><td>4 min</td><td>17%</td></tr></tbody></table></div>
+        <h3>Coverage</h3><p><code>AGENTS.md</code> · checked<br /><code>aidd_docs/memory/testing.md</code> · not reviewed<br /><code>skill export</code> · missing</p>
+        <details class="source-records"><summary>Raw sources</summary><p><code>run_2026-09-12_01</code> · <code>aidd telemetry report v8</code> · 2026-09-12T02:00Z → 03:00Z</p></details>
       </details>
 
       <footer class="prompt-builder" aria-labelledby="prompt-title">
-        <div class="prompt-heading"><h2 id="prompt-title">Prompt d’exécution</h2><span id="accepted-count" aria-live="polite">0 recommandation acceptée</span></div>
-        <textarea id="execution-prompt" rows="10" spellcheck="true">Applique uniquement les changements acceptés. Vérifie chaque source, préserve les changements existants, puis valide.
+        <div class="prompt-heading"><h2 id="prompt-title">Execution prompt</h2><span id="accepted-count" aria-live="polite">0 recommendations accepted</span></div>
+        <textarea id="execution-prompt" rows="10" spellcheck="true">Apply only the accepted changes. Verify each source, preserve existing changes, then validate.
 
-Changements acceptés :
+Accepted changes:
 
-Après validation, propose un petit changement général pour améliorer le prochain run et mesurer son effet.</textarea>
-        <div class="prompt-actions"><span>Le texte reste éditable.</span><button class="button" id="copy-prompt" type="button">Copier le prompt</button></div>
+---</textarea>
+        <div class="prompt-actions"><span>The text remains editable.</span><button class="button" id="copy-prompt" type="button" data-label-success="Prompt copied" data-label-error="Text selected">Copy prompt</button></div>
       </footer>
     </main>
   </body>
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report.js b/plugins/aidd-refine/skills/05-improve/assets/report.js
index f36fda79e..0b91b3bec 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report.js
+++ b/plugins/aidd-refine/skills/05-improve/assets/report.js
@@ -1,5 +1,10 @@
 (() => {
-  const promptAnchor = "Changements acceptés :";
+  const promptMarker = "---";
+
+  function formatCount(count, one, other) {
+    const template = count === 1 ? one : other;
+    return template ? template.replace("{count}", String(count)) : String(count);
+  }
 
   function composePrompt(value, accepted, savedLines = new Map(), knownIds = new Set(accepted.map(({ id }) => id))) {
     const generatedPattern = /^- \[([^\]]+)\] /;
@@ -15,15 +20,16 @@
       return !match || !knownIds.has(match[1]);
     });
     const generated = accepted.map(({ id, prompt }) => savedLines.get(id) || `- [${id}] ${prompt}`);
-    const anchorIndex = lines.findIndex((line) => line.trim() === promptAnchor);
-    lines.splice(anchorIndex >= 0 ? anchorIndex + 1 : lines.length, 0, ...generated);
+    const markerIndex = lines.findIndex((line) => line.trim() === promptMarker);
+    lines.splice(markerIndex >= 0 ? markerIndex : lines.length, 0, ...generated);
     return lines.join("\n");
   }
 
-  if (typeof module !== "undefined") module.exports = { composePrompt };
+  if (typeof module !== "undefined") module.exports = { composePrompt, formatCount };
   if (typeof document === "undefined") return;
 
   const state = { signal: "all", target: "all" };
+  const report = document.querySelector("#report");
   const findings = [...document.querySelectorAll(".finding")];
   const count = document.querySelector("#result-count");
   const empty = document.querySelector("#empty-state");
@@ -40,7 +46,7 @@
       finding.hidden = !(matchesSignal && matchesTarget);
       if (!finding.hidden) visible += 1;
     }
-    count.textContent = `${visible} recommandation${visible === 1 ? "" : "s"}`;
+    count.textContent = formatCount(visible, report.dataset.labelResultOne, report.dataset.labelResultOther);
     empty.hidden = visible !== 0;
   }
 
@@ -60,9 +66,9 @@
       const original = button.textContent;
       try {
         await navigator.clipboard.writeText(button.dataset.copy);
-        button.textContent = "Copié";
+        button.textContent = button.dataset.labelSuccess || original;
       } catch {
-        button.textContent = "Copie indisponible";
+        button.textContent = button.dataset.labelError || original;
       }
       window.setTimeout(() => { button.textContent = original; }, 1400);
     });
@@ -77,14 +83,15 @@
     prompt.value = composePrompt(prompt.value, entries, savedPromptLines, findingIds);
 
     const total = accepted.length;
-    acceptedCount.textContent = `${total} recommandation${total === 1 ? "" : "s"} acceptée${total === 1 ? "" : "s"}`;
+    acceptedCount.textContent = formatCount(total, report.dataset.labelAcceptedOne, report.dataset.labelAcceptedOther);
   }
 
   document.querySelectorAll(".accept-input").forEach((input) => {
     input.addEventListener("change", () => {
       const finding = input.closest(".finding");
       finding.classList.toggle("is-accepted", input.checked);
-      finding.querySelector(".accept-toggle span").textContent = input.checked ? "Accepté" : "Accepter";
+      const label = finding.querySelector(".accept-toggle span");
+      label.textContent = input.checked ? input.dataset.labelOn : input.dataset.labelOff;
       updatePrompt();
     });
   });
@@ -94,11 +101,11 @@
     const original = button.textContent;
     try {
       await navigator.clipboard.writeText(prompt.value);
-      button.textContent = "Prompt copié";
+      button.textContent = button.dataset.labelSuccess || original;
     } catch {
       prompt.focus();
       prompt.select();
-      button.textContent = "Texte sélectionné";
+      button.textContent = button.dataset.labelError || original;
     }
     window.setTimeout(() => { button.textContent = original; }, 1600);
   });
diff --git a/scripts/__tests__/improve-report.test.js b/scripts/__tests__/improve-report.test.js
index 374699090..1eb1f41a1 100644
--- a/scripts/__tests__/improve-report.test.js
+++ b/scripts/__tests__/improve-report.test.js
@@ -2,7 +2,7 @@ const assert = require('node:assert/strict');
 const { resolve } = require('node:path');
 const test = require('node:test');
 
-const { composePrompt } = require(resolve(
+const { composePrompt, formatCount } = require(resolve(
   __dirname,
   '../../plugins/aidd-refine/skills/05-improve/assets/report.js',
 ));
@@ -13,9 +13,11 @@ test('accepted edits keep report order and preserve manual prompt text', () => {
   const base = [
     'User introduction.',
     '',
-    'Changements acceptés :',
+    'Accepted changes:',
     '- [NOTE] Keep this manual note.',
     '',
+    '---',
+    '',
     'User closing.',
   ].join('\n');
 
@@ -31,6 +33,7 @@ test('accepted edits keep report order and preserve manual prompt text', () => {
   assert.match(value, /^User introduction\./);
   assert.match(value, /- \[NOTE\] Keep this manual note\./);
   assert.match(value, /User closing\.$/);
+  assert.ok(value.indexOf('[F-002]') < value.indexOf('---'));
 
   value = composePrompt(value, [{ id: 'F-001', prompt: 'First edit.' }], saved, known);
   assert.doesNotMatch(value, /\[F-002\]/);
@@ -44,3 +47,31 @@ test('accepted edits keep report order and preserve manual prompt text', () => {
     { id: 'F-002', prompt: 'Second edit.' },
   ], saved, known), value);
 });
+
+test('prompt insertion is independent of localized headings', () => {
+  for (const [heading, closing] of [
+    ['Accepted changes:', 'After validation, propose one improvement.'],
+    ['Changements acceptés :', 'Après validation, propose une amélioration.'],
+  ]) {
+    const base = ['Introduction.', '', heading, '', '---', '', closing].join('\n');
+    const value = composePrompt(base, [{ id: 'F-001', prompt: 'Apply the edit.' }]);
+
+    assert.ok(value.indexOf('[F-001]') < value.indexOf('---'));
+    assert.ok(value.indexOf('---') < value.indexOf(closing));
+    assert.match(value, new RegExp(heading.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')));
+  }
+});
+
+test('missing prompt marker preserves manual text and appends accepted edits safely', () => {
+  const base = ['Manual introduction.', '', 'Manual closing.'].join('\n');
+  const value = composePrompt(base, [{ id: 'F-001', prompt: 'Apply the edit.' }]);
+
+  assert.match(value, /^Manual introduction\./);
+  assert.match(value, /Manual closing\./);
+  assert.ok(value.indexOf('Manual closing.') < value.indexOf('[F-001]'));
+});
+
+test('count labels come from localized HTML data', () => {
+  assert.equal(formatCount(1, '{count} recommendation', '{count} recommendations'), '1 recommendation');
+  assert.equal(formatCount(2, '{count} recommandation', '{count} recommandations'), '2 recommandations');
+});

From e8a3434729ca5e19e37c7bc655fcce18975c6205 Mon Sep 17 00:00:00 2001
From: alexsoyes <contact.alexsoyes@gmail.com>
Date: Wed, 7 Oct 2026 10:07:00 +0700
Subject: [PATCH 7/7] fix(aidd-refine): streamline improve analysis and report

Keep optimization analysis focused and reports directly actionable.
---
 .../actions/01-read-conversation.md           |  10 +-
 .../skills/05-improve/actions/02-recommend.md |  20 ++-
 .../05-improve/actions/03-target-edits.md     |  18 +--
 .../05-improve/assets/report-template.html    | 130 ++++++++++--------
 .../skills/05-improve/assets/report.css       | 129 ++++++++---------
 .../skills/05-improve/assets/report.js        |  45 +-----
 scripts/__tests__/improve-report.test.js      |  99 +++++++++++++
 7 files changed, 259 insertions(+), 192 deletions(-)

diff --git a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
index ede26347b..02a09fa04 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/01-read-conversation.md
@@ -8,16 +8,22 @@ A current conversation, an exact conversation ID, or a complete transcript expor
 
 ## Output
 
-An evidence boundary, a `## Timing` table with `Activity | Observed time | Share | Evidence`, a `## Usage` table with `Metric | Value | Evidence`, and a scope index of encountered resources.
+- Evidence boundary and resource index.
+- Chronological timeline with stable event IDs: `Moment | Actor/tool | Action/source | Observed duration | Result`.
+- Timing: `Activity | Observed time | Share | Evidence`.
+- Usage: `Metric | Value | Evidence`.
 
 ## Process
 
 1. **Resolve.** Use the host route in [conversation sources](../assets/conversation-sources.md). Prefer its complete export or host reader. Match the exact session ID before direct storage; search only for that ID, never unrelated sessions, configuration, or authentication. Stop unless the exact complete transcript is available.
 2. **Freeze.** End before this invocation, or at the export boundary. Exclude every `improve` invocation and report.
 3. **Read once.** Load every in-scope message, tool call, result, and timestamp. Index relevant turns; invoked skills and their used actions or resources; applicable `AGENTS.md`; and task-relevant memory references. Do not read maintained sources yet.
+   - Record one timeline row per message, call, or result, including failures and retries. Link call/result pairs; retain full text, arguments, and output.
+   - Preserve transcript order. Identify agents, linked calls, and parallel activity when exposed; do not invent hidden calls.
 4. **Measure.** Use only timestamps, elapsed records, and exposed host usage.
    - Group tool time as `research and diagnosis`, `implementation`, `validation`, or `unattributed`.
+   - Attribute each call duration once; never sum overlapping intervals into elapsed time. Mark unknown durations `unavailable`.
    - Include tokens, requests, and cost only when exposed; otherwise mark them `unavailable`.
    - Separate background-process lifetime from blocking time.
    - Treat private reasoning duration as unavailable unless exposed directly; never infer it.
-5. **Deliver.** Order known activity times descending and cite every value.
+5. **Deliver.** Cite every measured value. Order only activity totals by descending time; keep the timeline chronological.
diff --git a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
index 9180a14d0..49d54832a 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/02-recommend.md
@@ -4,11 +4,11 @@ Reduce avoidable work without reducing reliability.
 
 ## Input
 
-The evidence boundary, timing and usage tables, complete conversation, and resource index.
+The evidence boundary, timeline, timing and usage tables, complete conversation, and resource index.
 
 ## Output
 
-An updated `Scope | Status | Evidence` index, a `## Recommendations` table with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Source | Saving`, and non-actionable limitations.
+Coverage with `Scope | Status | Evidence`, and recommendations with `ID | Focus | Type | Diagnostic | Evidence | Smallest change | Source | Saving`. Pass through the boundary, timeline, and measurements unchanged.
 
 ## Process
 
@@ -18,20 +18,16 @@ An updated `Scope | Status | Evidence` index, a `## Recommendations` table with
    - Prefer a cheap model and low reasoning effort when per-agent overrides are supported; otherwise inherit host defaults.
    - Give the behavior analyst the frozen transcript. Give artifact analysts the boundary, indexed turns, and verified excerpts with evidence; allow a targeted read-only refresh only when validity is uncertain.
    - Give every analyst steps 4–6 and require table rows.
-4. **Reflect.** Ask yourself, for each scope; keep this assessment internal:
-   - How could the next run be faster or better?
-   - What should be removed or clarified? What was counterproductive?
-   - Where should change live? How could it save time or tokens?
-   - Which back-and-forth, bottlenecks, or tool calls could be removed, batched, parallelized, or replaced?
+4. **Reflect.** Ask yourself one question per scope, internally: "Based on the conversation, what shorter, equally reliable path could have achieved the same result, and what minimal reusable change would reduce time, tokens, or cost next run?"
 5. **Assess.** Use `obsolete`, `over-specific-or-time-bound`, `duplicate`, `inconsistent`, `counterproductive`, or `correct`.
-   - Compare observed work with the shortest reliable path to the requested result. Keep checks for mutable state or unresolved uncertainty; reuse still-valid results instead of repeating research.
+   - Assess repeated reads, back-and-forth, bottlenecks, and calls to remove, batch, parallelize, or replace. Keep checks for mutable state or unresolved uncertainty; reuse valid results.
    - Raw call count never proves waste. Ground the smallest general correction in evidence.
    - Recommend memory only for concise, durable knowledge that prevents recurring rediscovery, never transient state or one-off implementation details.
    - For skills, especially knowledge: delete evidence-backed waste or inconsistency first, without quota; then consolidate or clarify.
    - Reuse or strengthen existing content; add only for a demonstrated gap it cannot cover. Preserve useful context and requirements; classify sufficient content as `correct` => `no change`.
-   - Claim time savings from blocking-time or resource-impact evidence, never background-process uptime alone.
+   - Ground savings in blocking-time, resource-impact, or exposed cost evidence. State trade-offs; background-process uptime alone proves no saving.
 6. **Deliver.** Deduplicate, verify cited evidence, and order by focus, then `behavior`, `skill`, `knowledge`. Report findings, not the questionnaire, in the fewest actionable words.
-   - Every recommendation needs an exact maintained source path, current excerpt, minimal correction or explicit `+`/`−` edit, and a short target/action prompt.
-   - If no source resolves or no useful persistent edit exists, report a non-actionable limitation.
+   - Every recommendation needs an exact maintained source path, current excerpt, explicit `−`/`+` correction, relevant event IDs, and a short target/action prompt. Leave `+` empty for deletion-only edits.
+   - Disclose unresolved sources in coverage. Omit findings without a useful persistent edit; do not invent recommendations.
    - Type: `skill`, `behavior`, `knowledge`, or `tooling`.
-   - Saving: `time`, `tokens`, `both`, or `unknown`.
+   - Saving: `time`, `tokens`, `cost`, their combination, or `unknown`.
diff --git a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
index ec8f9c812..aa8dbe195 100644
--- a/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
+++ b/plugins/aidd-refine/skills/05-improve/actions/03-target-edits.md
@@ -1,10 +1,10 @@
 # 03 - Target edits
 
-Map minimal edits and render the report.
+Render the evidence and minimal edits.
 
 ## Input
 
-The timing, usage, recommendations, and scope index tables.
+The evidence boundary, timeline, timing, usage, coverage, and recommendations.
 
 ## Output
 
@@ -12,9 +12,11 @@ An HTML report in a temporary directory with local `report.css` and `report.js`,
 
 ## Process
 
-1. **Target.** Consolidate the provided recommendations by source into `File | + Added | − Removed or clarified`; use `No recommended file | — | —` when empty. Never resolve, judge, or apply sources here.
-2. **Render.** Fill [the report template](../assets/report-template.html) from the provided data; remove its sample and copy its local CSS and JavaScript into the allowed unique temporary directory.
-   - Escape injected values; allow only template markup and local assets. Localize the HTML `lang`, visible text, and dynamic `data-label-*` text to the user's language.
-   - Put findings first with diagnostic/type tags, filters, accept controls, the target table, and an editable global prompt. Copy each recommendation's short target/action prompt into `data-prompt` without redrafting it.
-   - Show source paths and minimal edits. Collapse raw evidence, secondary metrics, timing, coverage, source records, and limitations; keep the run ID in source records and avoid repeated metadata.
-3. **Return.** Return its path. Ask the user, in their language, which small general change in intent to apply next run for cumulative, measurable improvement.
+1. **Render.** Fill [the report template](../assets/report-template.html) from the provided data; remove its sample and copy its local CSS and JavaScript into the allowed unique temporary directory. Never resolve, judge, or apply sources here.
+   - Escape injected values; allow only template markup and local assets. Localize HTML `lang`, UI text, and dynamic `data-label-*` text; preserve quoted evidence.
+   - Start with elapsed seconds, tokens, and call counts only. Use exposed usage only; distinguish linked subcalls. No visible dates or coverage. Keep text sizes uniform except titles and user prompts.
+   - Show full user prompts in labelled blockquotes, followed by compact calls: tool, relative path or command, and elapsed gap since the previous displayed event. Nest only exposed subcalls; preserve chronological order and parallel relationships.
+   - Derive gaps only from recorded timestamps; distinguish them from call durations. Omit assistant prose, raw payloads, and evidence explanations. No dropdowns.
+   - Put recommendations below: an actionable title, one relative path, and unlabelled red `−` / green `+` columns. Use `—` for an empty side; omit descriptions and proof links.
+   - Keep black emoji tags, accept controls, and the editable prompt; copy each `data-prompt` unchanged. No filters, print control, target table, or limitations section.
+2. **Return.** Return its path. Ask the user, in their language, which small general change in intent to apply next run for cumulative, measurable improvement.
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report-template.html b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
index bc239f99e..d571102aa 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report-template.html
+++ b/plugins/aidd-refine/skills/05-improve/assets/report-template.html
@@ -11,88 +11,108 @@
   <body>
     <header class="site-header">
       <a class="brand" href="#report" aria-label="AIDD Improve report"><span class="brand-mark" aria-hidden="true">A</span><span>AIDD / improve</span></a>
-      <div class="header-actions"><button class="button" id="print-report" type="button">Print</button></div>
     </header>
 
-    <main id="report" data-label-result-one="{count} recommendation" data-label-result-other="{count} recommendations" data-label-accepted-one="{count} recommendation accepted" data-label-accepted-other="{count} recommendations accepted">
+    <main id="report" data-label-accepted-one="{count} recommendation accepted" data-label-accepted-other="{count} recommendations accepted">
       <section class="intro" aria-labelledby="report-title">
         <h1 id="report-title">Improve report</h1>
-        <p>Example only · fictional content · partial coverage · 1 recommendation</p>
+        <p>Fictional example.</p>
       </section>
 
-      <section class="section findings-section" aria-labelledby="findings-title">
-        <div class="section-heading"><h2 id="findings-title">Recommended changes</h2><p id="result-count" class="result-count" aria-live="polite">1 recommendation</p></div>
+      <section class="recap" aria-label="Conversation recap" data-start="2026-09-12T09:00:00Z" data-end="2026-09-12T09:00:09Z">
+        <div class="metrics">
+          <p><span>Elapsed</span><strong>9 s</strong></p>
+          <p><span>Tokens</span><strong>Unavailable</strong></p>
+          <p><span>Calls · includes 2 subcalls</span><strong>6</strong></p>
+        </div>
+      </section>
 
-        <details class="filter-panel">
-          <summary>Filters</summary>
-          <div class="filters" aria-label="Filter recommendations">
-            <div class="filter-group" role="group" aria-label="Diagnostic">
-              <span>Diagnostic</span>
-              <button type="button" data-filter="signal" data-value="all" aria-pressed="true">All</button>
-              <button type="button" data-filter="signal" data-value="obsolete" aria-pressed="false">Obsolete</button>
-              <button type="button" data-filter="signal" data-value="over-specific-or-time-bound" aria-pressed="false">Over-specific</button>
-              <button type="button" data-filter="signal" data-value="duplicate" aria-pressed="false">Duplicate</button>
-              <button type="button" data-filter="signal" data-value="inconsistent" aria-pressed="false">Inconsistent</button>
-              <button type="button" data-filter="signal" data-value="counterproductive" aria-pressed="false">Counterproductive</button>
-            </div>
-            <div class="filter-group" role="group" aria-label="Target type">
-              <span>Target</span>
-              <button type="button" data-filter="target" data-value="all" aria-pressed="true">All</button>
-              <button type="button" data-filter="target" data-value="behavior" aria-pressed="false">Behavior</button>
-              <button type="button" data-filter="target" data-value="skill" aria-pressed="false">Skill</button>
-              <button type="button" data-filter="target" data-value="knowledge" aria-pressed="false">Knowledge</button>
-              <button type="button" data-filter="target" data-value="tooling" aria-pressed="false">Tooling</button>
-            </div>
+      <section class="section conversation-section" aria-labelledby="conversation-title">
+        <div class="section-heading"><h2 id="conversation-title">Conversation</h2></div>
+
+        <article class="conversation-turn">
+          <blockquote class="user-message" data-moment="2026-09-12T09:00:00Z" aria-labelledby="first-prompt-label">
+            <div class="message-meta"><span id="first-prompt-label">User prompt</span><span class="event-delta">Start</span></div>
+            <p class="user-prompt">Review the project guidance for repetition and unnecessary output. Suggest changes only.</p>
+          </blockquote>
+
+          <div class="tool-calls" aria-label="Calls following the first prompt">
+            <article class="tool-call" data-call-id="E003" data-moment="2026-09-12T09:00:02Z">
+              <div class="tool-line"><span class="tool-name">read_file</span><code class="tool-target">AGENTS.md</code><span class="event-delta" aria-label="2 seconds since previous displayed item">+2 s</span></div>
+            </article>
+            <article class="tool-call" data-call-id="E006" data-moment="2026-09-12T09:00:04Z">
+              <div class="tool-line"><span class="tool-name">parallel</span><span class="tool-target">2 linked read_file subcalls</span><span class="event-delta" aria-label="2 seconds since previous displayed item">+2 s</span></div>
+              <div class="subcalls" aria-label="Linked parallel subcalls">
+                <article class="tool-call" data-call-id="E007" data-parent="E006" data-moment="2026-09-12T09:00:04.100Z">
+                  <div class="tool-line"><span class="tool-name">read_file</span><code class="tool-target">docs/example-review.md</code><span class="event-delta" aria-label="0.1 seconds since previous displayed item">+0.1 s</span></div>
+                </article>
+                <article class="tool-call" data-call-id="E008" data-parent="E006" data-moment="2026-09-12T09:00:04.200Z">
+                  <div class="tool-line"><span class="tool-name">read_file</span><code class="tool-target">docs/example-guidance.md</code><span class="event-delta" aria-label="0.1 seconds since previous displayed item">+0.1 s</span></div>
+                </article>
+              </div>
+            </article>
+            <article class="tool-call" data-call-id="E012" data-moment="2026-09-12T09:00:05Z">
+              <div class="tool-line"><span class="tool-name">read_file</span><span class="tool-target"><code>docs/example-review-notes.md</code> <span class="call-error">ENOENT</span></span><span class="event-delta" aria-label="0.8 seconds since previous displayed item">+0.8 s</span></div>
+            </article>
+            <article class="tool-call" data-call-id="E014" data-moment="2026-09-12T09:00:06Z">
+              <div class="tool-line"><span class="tool-name">shell</span><span class="tool-target"><code>rg --files docs -g '*review*'</code> <span class="call-status">retry</span></span><span class="event-delta" aria-label="1 second since previous displayed item">+1 s</span></div>
+            </article>
           </div>
-        </details>
+        </article>
+
+        <article class="conversation-turn">
+          <blockquote class="user-message" data-moment="2026-09-12T09:00:08Z" aria-labelledby="second-prompt-label">
+            <div class="message-meta"><span id="second-prompt-label">User prompt</span><span class="event-delta" aria-label="2 seconds since previous displayed item">+2 s</span></div>
+            <p class="user-prompt">Show replacements and deletions side by side. Do not invent a new rule for the deleted duplicate.</p>
+          </blockquote>
+        </article>
+      </section>
+
+      <section class="section findings-section" aria-labelledby="findings-title">
+        <div class="section-heading"><h2 id="findings-title">Recommended changes</h2></div>
 
         <div class="findings" id="findings-list">
-          <article class="finding" data-signal="duplicate" data-target="knowledge">
+          <article class="finding" data-id="concise-output" data-signal="counterproductive" data-target="behavior">
+            <header class="finding-header">
+              <div class="finding-title"><div class="tags"><span class="tag"><span aria-hidden="true">⚠️</span> Counterproductive</span><span class="tag"><span aria-hidden="true">🧭</span> Behavior</span></div><h3>Replace the full-output requirement</h3></div>
+              <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="In docs/example-review.md, replace 'Before replying, print every tool result in full.' with 'Report the result and the shortest evidence needed to verify it.'" data-label-off="Accept" data-label-on="Accepted"><span>Accept</span></label>
+            </header>
+            <div class="finding-collapse"><div class="finding-body">
+              <div class="change">
+                <div class="change-header"><code>docs/example-review.md</code></div>
+                <div class="diff-columns">
+                  <div class="diff-pane diff-del" aria-label="Removed or replaced"><pre class="diff"><code><span class="diff-line"><b>−</b>Before replying, print every tool result in full.</span></code></pre></div>
+                  <div class="diff-pane diff-add" aria-label="Added or kept"><pre class="diff"><code><span class="diff-line"><b>+</b>Report the result and the shortest evidence needed to verify it.</span></code></pre></div>
+                </div>
+              </div>
+            </div></div>
+          </article>
+          <article class="finding" data-id="duplicate-publishing-rule" data-signal="duplicate" data-target="knowledge">
             <header class="finding-header">
-              <div class="finding-id">F-001</div>
-              <div class="finding-title"><div class="tags"><span class="tag tag-signal">Duplicate</span><span class="tag">Knowledge</span></div><h3>Example · remove a fictional duplicate rule</h3></div>
+              <div class="finding-title"><div class="tags"><span class="tag"><span aria-hidden="true">♻️</span> Duplicate</span><span class="tag"><span aria-hidden="true">📚</span> Knowledge</span></div><h3>Remove the second publishing rule</h3></div>
               <label class="accept-toggle"><input class="accept-input" type="checkbox" data-prompt="In docs/example-guidance.md, remove the second duplicate 'Validate before publishing.' rule." data-label-off="Accept" data-label-on="Accepted"><span>Accept</span></label>
             </header>
             <div class="finding-collapse"><div class="finding-body">
               <div class="change">
-                <div class="change-header"><span>docs/example-guidance.md</span><button class="copy-path" type="button" data-copy="docs/example-guidance.md" data-label-success="Copied" data-label-error="Copy unavailable">Copy</button></div>
-                <pre class="diff" aria-label="Proposed change for example-guidance.md"><code><span class="diff-line"><b> </b>Validate before publishing.</span>
-<span class="diff-line diff-del"><b>−</b>Validate before publishing.</span></code></pre>
+                <div class="change-header"><code>docs/example-guidance.md</code></div>
+                <div class="diff-columns">
+                  <div class="diff-pane diff-del" aria-label="Removed or replaced"><pre class="diff"><code><span class="diff-line"><b>−</b>Validate before publishing.</span></code></pre></div>
+                  <div class="diff-pane diff-add" aria-label="Added or kept"><p class="diff-empty" aria-label="No addition">—</p></div>
+                </div>
               </div>
-              <p class="why">One rule carries the same requirement.</p>
-              <details class="evidence-details"><summary>Evidence</summary><p>The fictional source contains the same rule twice.</p></details>
             </div></div>
           </article>
-
         </div>
-        <p class="empty-state" id="empty-state" hidden>No recommendations match.</p>
-
-        <details class="limitations"><summary>Non-actionable limitations · 1</summary><p>A cited source could not be resolved, so no change is proposed.</p></details>
       </section>
 
-      <section class="section target-section" aria-labelledby="target-title">
-        <div class="section-heading"><h2 id="target-title">Target files</h2><span class="section-total">1 file</span></div>
-        <div class="table-wrap"><table><thead><tr><th>File</th><th>+ Added</th><th>− Removed or clarified</th></tr></thead><tbody>
-          <tr><td><code>docs/example-guidance.md</code></td><td>—</td><td>Second duplicate rule</td></tr>
-        </tbody></table></div>
-      </section>
-
-      <details class="section raw-section">
-        <summary>Metrics, timeline, and coverage</summary>
-        <div class="metrics"><p><span>Active time</span><strong>24 min</strong></p><p><span>Tokens</span><strong>128.4k</strong></p><p><span>Requests</span><strong>42</strong></p><p><span>Cost</span><strong>Unknown</strong></p></div>
-        <div class="table-wrap"><table><thead><tr><th>Activity</th><th>Time</th><th>Share</th></tr></thead><tbody><tr><td>Research and diagnosis</td><td>12 min</td><td>50%</td></tr><tr><td>Implementation</td><td>8 min</td><td>33%</td></tr><tr><td>Validation</td><td>4 min</td><td>17%</td></tr></tbody></table></div>
-        <h3>Coverage</h3><p><code>AGENTS.md</code> · checked<br /><code>aidd_docs/memory/testing.md</code> · not reviewed<br /><code>skill export</code> · missing</p>
-        <details class="source-records"><summary>Raw sources</summary><p><code>run_2026-09-12_01</code> · <code>aidd telemetry report v8</code> · 2026-09-12T02:00Z → 03:00Z</p></details>
-      </details>
-
       <footer class="prompt-builder" aria-labelledby="prompt-title">
         <div class="prompt-heading"><h2 id="prompt-title">Execution prompt</h2><span id="accepted-count" aria-live="polite">0 recommendations accepted</span></div>
-        <textarea id="execution-prompt" rows="10" spellcheck="true">Apply only the accepted changes. Verify each source, preserve existing changes, then validate.
+        <textarea id="execution-prompt" aria-labelledby="prompt-title" rows="10" spellcheck="true">Apply only the accepted changes. Verify each source, preserve existing changes, then validate.
 
 Accepted changes:
 
 ---</textarea>
-        <div class="prompt-actions"><span>The text remains editable.</span><button class="button" id="copy-prompt" type="button" data-label-success="Prompt copied" data-label-error="Text selected">Copy prompt</button></div>
+        <div class="prompt-actions"><button class="button" id="copy-prompt" type="button" data-label-success="Prompt copied" data-label-error="Text selected">Copy prompt</button></div>
       </footer>
     </main>
   </body>
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report.css b/plugins/aidd-refine/skills/05-improve/assets/report.css
index b7da388d5..2fea03391 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report.css
+++ b/plugins/aidd-refine/skills/05-improve/assets/report.css
@@ -3,11 +3,10 @@
   --red-dark: color-mix(in oklch, var(--red) 72%, black);
   --red-soft: color-mix(in oklch, var(--red) 12%, white);
   --green: #66cc99;
-  --green-dark: color-mix(in oklch, var(--green) 62%, black);
+  --green-dark: color-mix(in oklch, var(--green) 58%, black);
   --green-soft: color-mix(in oklch, var(--green) 15%, white);
   --ink: oklch(0.16 0.01 25);
   --soft-ink: oklch(0.4 0.01 25);
-  --muted: oklch(0.55 0.008 25);
   --line: oklch(0.87 0.006 25);
   --strong-line: oklch(0.72 0.01 25);
   --surface: white;
@@ -17,106 +16,92 @@
 }
 
 * { box-sizing: border-box; }
-body { margin: 0; background: var(--surface); color: var(--ink); font: 14px/1.6 var(--mono); }
+body { margin: 0; background: var(--surface); color: var(--ink); font: 13px/1.6 var(--mono); }
 button, textarea { font: inherit; }
-button:focus-visible, summary:focus-visible, textarea:focus { outline: 3px solid var(--focus); outline-offset: 3px; }
+button:focus-visible, a:focus-visible, textarea:focus { outline: 3px solid var(--focus); outline-offset: 3px; }
 code { font-family: var(--mono); }
+a { color: var(--red-dark); text-underline-offset: 3px; }
 
-.site-header { min-height: 58px; padding: 0 32px; border-bottom: 1px solid var(--line); display: flex; align-items: center; justify-content: space-between; position: sticky; top: 0; z-index: 10; background: color-mix(in oklch, white 94%, transparent); backdrop-filter: blur(12px); }
+.site-header { min-height: 58px; padding: 0 32px; border-bottom: 1px solid var(--line); display: flex; align-items: center; position: sticky; top: 0; z-index: 10; background: color-mix(in oklch, white 94%, transparent); backdrop-filter: blur(12px); }
 .brand { color: var(--ink); text-decoration: none; display: flex; align-items: center; gap: 10px; font-weight: 700; }
 .brand-mark { width: 25px; height: 25px; display: grid; place-items: center; background: var(--red); color: var(--ink); }
 .button { border: 1px solid var(--ink); padding: 7px 12px; background: var(--ink); color: white; cursor: pointer; }
 .button:hover { border-color: var(--red); background: var(--red); color: var(--ink); }
 
 main { width: min(1040px, calc(100% - 48px)); margin: auto; }
-.intro { padding: 58px 0 30px; border-bottom: 1px solid var(--ink); }
-.intro h1 { margin: 0; font-size: clamp(34px, 6vw, 48px); line-height: 1.05; letter-spacing: -0.045em; }
-.intro p { margin: 16px 0 0; color: var(--muted); font-size: 12px; }
-.section { padding: 42px 0; border-bottom: 1px solid var(--line); }
+.intro { padding: 24px 0 16px; }
+.intro h1 { margin: 0; font-size: 28px; line-height: 1.2; letter-spacing: -0.03em; }
+.intro p { margin: 8px 0 0; color: var(--soft-ink); }
+.section { padding: 24px 0; border-bottom: 1px solid var(--line); }
 .section-heading, .prompt-heading { display: flex; align-items: end; justify-content: space-between; gap: 20px; margin-bottom: 20px; }
-.section-heading h2, .prompt-heading h2 { margin: 0; font-size: 22px; letter-spacing: -0.03em; }
-.result-count, .section-total, .prompt-heading span, .prompt-actions span { color: var(--muted); font-size: 11px; }
+.section-heading h2, .prompt-heading h2 { margin: 0; font-size: 18px; letter-spacing: -0.03em; }
+.prompt-heading span { color: var(--soft-ink); }
 
-details > summary { cursor: pointer; color: var(--soft-ink); font-weight: 700; }
-.filter-panel { margin-bottom: 18px; border-block: 1px solid var(--line); padding: 11px 0; }
-.filters { display: flex; flex-wrap: wrap; gap: 12px 30px; padding-top: 12px; }
-.filter-group { display: flex; flex-wrap: wrap; align-items: center; gap: 6px; }
-.filter-group > span { margin-right: 4px; color: var(--muted); font-size: 11px; }
-.filter-group button { border: 1px solid var(--strong-line); border-radius: 99px; padding: 4px 8px; background: white; color: var(--soft-ink); font-size: 11px; cursor: pointer; }
-.filter-group button[aria-pressed="true"] { border-color: var(--ink); background: var(--ink); color: white; }
+.recap { padding-bottom: 16px; border-bottom: 1px solid var(--line); }
+.metrics { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); border: 1px solid var(--line); font-variant-numeric: tabular-nums; }
+.metrics p { margin: 0; padding: 12px 16px; display: grid; gap: 4px; align-content: start; overflow-wrap: anywhere; }
+.metrics p + p { border-left: 1px solid var(--line); }
+.metrics span { color: var(--soft-ink); }
+
+.conversation-turn + .conversation-turn { margin-top: 30px; }
+.user-message { margin: 0 0 16px; padding: 16px; border-left: 4px solid var(--ink); background: #f2f2f2; }
+.message-meta { display: flex; flex-wrap: wrap; align-items: center; gap: 12px; color: var(--soft-ink); font-variant-numeric: tabular-nums; }
+.message-meta > span:first-child { color: var(--ink); font-weight: 700; }
+.message-meta .event-delta { margin-left: auto; }
+.user-prompt { margin: 8px 0 0; font-size: 18px; line-height: 1.45; letter-spacing: -0.02em; }
+.tool-calls { padding-left: 28px; }
+.tool-call { margin: 12px 0; }
+.tool-line { display: grid; grid-template-columns: 92px minmax(0, 1fr) auto; align-items: baseline; gap: 10px; padding: 4px 0; }
+.tool-name { color: var(--soft-ink); font-weight: 700; }
+.tool-target { overflow-wrap: anywhere; }
+.event-delta { color: var(--soft-ink); white-space: nowrap; font-variant-numeric: tabular-nums; }
+.call-error { color: var(--red-dark); background: var(--red-soft); padding: 1px 4px; }
+.call-status { color: var(--soft-ink); }
+.subcalls { margin-left: 18px; padding-left: 16px; border-left: 1px solid var(--line); }
 
 .findings { border-top: 1px solid var(--ink); }
 .finding { border-bottom: 1px solid var(--ink); }
-.finding[hidden] { display: none; }
-.finding-header { padding: 22px 0; display: grid; grid-template-columns: 58px minmax(0, 1fr) auto; gap: 18px; align-items: start; }
-.finding-id { color: var(--red-dark); font-size: 11px; font-weight: 700; }
+.finding-header { padding: 22px 0; display: grid; grid-template-columns: minmax(0, 1fr) auto; gap: 18px; align-items: start; }
 .finding-title h3 { margin: 9px 0 0; font-size: 18px; line-height: 1.3; }
-.tags { display: flex; gap: 6px; }
-.tag { border: 1px solid var(--strong-line); padding: 3px 7px; font-size: 11px; }
-.tag-signal { border-color: var(--red); background: var(--red); color: var(--ink); }
-.accept-toggle { min-height: 31px; border: 1px solid var(--ink); padding: 5px 9px; display: flex; align-items: center; gap: 7px; font-size: 11px; font-weight: 700; cursor: pointer; }
+.tags { display: flex; flex-wrap: wrap; gap: 6px; }
+.tag { background: black; color: white; padding: 3px 7px; }
+.accept-toggle { min-height: 31px; border: 1px solid var(--ink); padding: 5px 9px; display: flex; align-items: center; gap: 7px; font-weight: 700; cursor: pointer; }
 .accept-input { width: 14px; height: 14px; margin: 0; accent-color: var(--red); }
 .finding-collapse { display: grid; grid-template-rows: 1fr; transition: grid-template-rows 180ms ease-out, opacity 180ms ease-out; }
-.finding-body { min-height: 0; padding: 0 0 26px 76px; overflow: hidden; }
+.finding-body { min-height: 0; padding-bottom: 26px; overflow: hidden; }
 .finding.is-accepted .finding-header { padding-block: 12px; opacity: 0.55; }
 .finding.is-accepted .finding-collapse { grid-template-rows: 0fr; opacity: 0.35; }
+.finding.is-accepted .finding-body { padding-block: 0; }
 .finding.is-accepted .accept-toggle { border-color: var(--red); background: var(--red-soft); color: var(--red-dark); }
 
-.change { border: 1px solid var(--strong-line); background: var(--soft-surface); overflow: hidden; }
-.change-header { min-height: 38px; padding: 8px 11px; border-bottom: 1px solid var(--line); display: flex; justify-content: space-between; gap: 16px; font-size: 11px; }
-.change-header span { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
-.copy-path { border: 0; padding: 0; background: transparent; color: var(--red-dark); font-size: 11px; cursor: pointer; }
-.diff { margin: 0; padding: 9px 0; overflow-x: auto; font: 11px/1.8 var(--mono); }
-.diff code { display: block; min-width: max-content; }
-.diff-line { display: block; padding: 0 13px; }
+.change { border: 1px solid var(--strong-line); overflow: hidden; }
+.change-header { padding: 8px 11px; border-bottom: 1px solid var(--line); }
+.change-header code { overflow-wrap: anywhere; }
+.diff-columns { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); }
+.diff-pane + .diff-pane { border-left: 1px solid var(--line); }
+.diff { margin: 0; padding: 11px 0; font: inherit; white-space: pre-wrap; overflow-wrap: anywhere; }
+.diff code { display: block; }
+.diff-line { display: block; padding: 0 11px; }
 .diff-line b { display: inline-block; width: 18px; user-select: none; }
 .diff-del { background: var(--red-soft); color: var(--red-dark); }
-.diff-add { background: var(--green-soft); color: var(--ink); }
-.why { margin: 12px 0; color: var(--soft-ink); font-size: 11px; }
-.evidence-details { margin-top: 10px; font-size: 11px; }
-.evidence-details p { margin: 8px 0 0; padding: 10px; background: var(--soft-surface); }
-.empty-state { padding: 35px 0; color: var(--muted); text-align: center; }
-.limitations { margin-top: 18px; padding: 12px 0; border-bottom: 1px solid var(--line); font-size: 11px; }
-
-.table-wrap { overflow-x: auto; border-top: 1px solid var(--ink); }
-table { width: 100%; min-width: 650px; border-collapse: collapse; font-size: 11px; text-align: left; }
-th { padding: 10px 12px; border-bottom: 1px solid var(--ink); color: var(--muted); font-size: 11px; font-weight: 500; }
-td { padding: 13px 12px; border-bottom: 1px solid var(--line); vertical-align: top; }
-th:first-child, td:first-child { padding-left: 0; }
-th:last-child, td:last-child { padding-right: 0; }
-td code { color: var(--red-dark); font-weight: 700; }
+.diff-add { background: var(--green-soft); color: var(--green-dark); }
+.diff-empty { margin: 0; padding: 11px; color: var(--soft-ink); }
 
-.raw-section { padding: 20px 0; }
-.raw-section > summary { font-size: 15px; }
-.raw-section[open] > summary { margin-bottom: 18px; }
-.metrics { display: grid; grid-template-columns: repeat(4, 1fr); border-top: 1px solid var(--ink); }
-.metrics p { margin: 0; padding: 14px; border-right: 1px solid var(--line); display: grid; gap: 5px; }
-.metrics p:last-child { border-right: 0; }
-.metrics span { color: var(--muted); font-size: 11px; }
-.metrics strong { font-size: 18px; }
-.raw-section h3 { margin: 20px 0 5px; font-size: 12px; }
-.raw-section p { color: var(--soft-ink); font-size: 11px; }
-.source-records { margin-top: 14px; font-size: 11px; }
-
-.prompt-builder { margin: 54px 0; padding-top: 24px; border-top: 4px solid var(--red); }
-#execution-prompt { width: 100%; min-height: 220px; padding: 15px; border: 1px solid var(--ink); resize: vertical; background: var(--soft-surface); color: var(--ink); font: 12px/1.7 var(--mono); }
-.prompt-actions { margin-top: 10px; display: flex; align-items: center; justify-content: space-between; gap: 20px; }
+.prompt-builder { margin: 28px 0; padding-top: 24px; border-top: 4px solid var(--red); }
+#execution-prompt { width: 100%; min-height: 220px; padding: 15px; border: 1px solid var(--ink); resize: vertical; background: var(--soft-surface); color: var(--ink); }
+.prompt-actions { margin-top: 10px; display: flex; justify-content: end; }
 
 @media (max-width: 680px) {
   .site-header { padding: 0 16px; }
   main { width: min(100% - 32px, 1040px); }
+  .metrics { grid-template-columns: 1fr; }
+  .metrics p + p { border-left: 0; border-top: 1px solid var(--line); }
+  .tool-calls { padding-left: 20px; }
+  .tool-line { grid-template-columns: 76px minmax(0, 1fr) auto; gap: 6px; }
+  .subcalls { margin-left: 8px; padding-left: 8px; }
   .finding-header { grid-template-columns: 1fr; gap: 8px; }
   .accept-toggle { justify-self: start; }
-  .finding-body { padding-left: 0; }
-  .metrics { grid-template-columns: repeat(2, 1fr); }
-  .metrics p:nth-child(2) { border-right: 0; }
-  .section-heading, .prompt-heading, .prompt-actions { align-items: start; flex-direction: column; }
+  .section-heading, .prompt-heading { align-items: start; flex-direction: column; }
 }
 
 @media (prefers-reduced-motion: reduce) { .finding-collapse { transition-duration: 0.001ms; } }
-@media print {
-  .site-header, .filter-panel, .copy-path, .accept-toggle, .prompt-actions { display: none !important; }
-  main { width: 100%; }
-  .finding[hidden] { display: block; }
-  .finding.is-accepted { display: none; }
-}
diff --git a/plugins/aidd-refine/skills/05-improve/assets/report.js b/plugins/aidd-refine/skills/05-improve/assets/report.js
index 0b91b3bec..a7801ce11 100644
--- a/plugins/aidd-refine/skills/05-improve/assets/report.js
+++ b/plugins/aidd-refine/skills/05-improve/assets/report.js
@@ -28,56 +28,17 @@
   if (typeof module !== "undefined") module.exports = { composePrompt, formatCount };
   if (typeof document === "undefined") return;
 
-  const state = { signal: "all", target: "all" };
   const report = document.querySelector("#report");
   const findings = [...document.querySelectorAll(".finding")];
-  const count = document.querySelector("#result-count");
-  const empty = document.querySelector("#empty-state");
   const prompt = document.querySelector("#execution-prompt");
   const acceptedCount = document.querySelector("#accepted-count");
   const savedPromptLines = new Map();
-  const findingIds = new Set(findings.map((finding) => finding.querySelector(".finding-id").textContent.trim()));
-
-  function render() {
-    let visible = 0;
-    for (const finding of findings) {
-      const matchesSignal = state.signal === "all" || finding.dataset.signal === state.signal;
-      const matchesTarget = state.target === "all" || finding.dataset.target === state.target;
-      finding.hidden = !(matchesSignal && matchesTarget);
-      if (!finding.hidden) visible += 1;
-    }
-    count.textContent = formatCount(visible, report.dataset.labelResultOne, report.dataset.labelResultOther);
-    empty.hidden = visible !== 0;
-  }
-
-  document.querySelectorAll("[data-filter]").forEach((button) => {
-    button.addEventListener("click", () => {
-      const kind = button.dataset.filter;
-      state[kind] = button.dataset.value;
-      document.querySelectorAll(`[data-filter="${kind}"]`).forEach((candidate) => {
-        candidate.setAttribute("aria-pressed", String(candidate === button));
-      });
-      render();
-    });
-  });
-
-  document.querySelectorAll(".copy-path").forEach((button) => {
-    button.addEventListener("click", async () => {
-      const original = button.textContent;
-      try {
-        await navigator.clipboard.writeText(button.dataset.copy);
-        button.textContent = button.dataset.labelSuccess || original;
-      } catch {
-        button.textContent = button.dataset.labelError || original;
-      }
-      window.setTimeout(() => { button.textContent = original; }, 1400);
-    });
-  });
+  const findingIds = new Set(findings.map((finding) => finding.dataset.id));
 
   function updatePrompt() {
     const accepted = findings.filter((finding) => finding.querySelector(".accept-input").checked);
     const entries = accepted.map((finding) => ({
-      id: finding.querySelector(".finding-id").textContent.trim(),
+      id: finding.dataset.id,
       prompt: finding.querySelector(".accept-input").dataset.prompt,
     }));
     prompt.value = composePrompt(prompt.value, entries, savedPromptLines, findingIds);
@@ -109,6 +70,4 @@
     }
     window.setTimeout(() => { button.textContent = original; }, 1600);
   });
-
-  document.querySelector("#print-report").addEventListener("click", () => window.print());
 })();
diff --git a/scripts/__tests__/improve-report.test.js b/scripts/__tests__/improve-report.test.js
index 1eb1f41a1..8a81779ea 100644
--- a/scripts/__tests__/improve-report.test.js
+++ b/scripts/__tests__/improve-report.test.js
@@ -1,6 +1,8 @@
 const assert = require('node:assert/strict');
+const { readFileSync } = require('node:fs');
 const { resolve } = require('node:path');
 const test = require('node:test');
+const { runInNewContext } = require('node:vm');
 
 const { composePrompt, formatCount } = require(resolve(
   __dirname,
@@ -75,3 +77,100 @@ test('count labels come from localized HTML data', () => {
   assert.equal(formatCount(1, '{count} recommendation', '{count} recommendations'), '1 recommendation');
   assert.equal(formatCount(2, '{count} recommandation', '{count} recommandations'), '2 recommandations');
 });
+
+function loadReport() {
+  function element(fields = {}) {
+    const listeners = new Map();
+    return {
+      ...fields,
+      addEventListener: (name, listener) => listeners.set(name, listener),
+      dispatch(name) { return listeners.get(name)({ currentTarget: this }); },
+    };
+  }
+
+  const prompt = element({ value: 'Manual introduction.\n\n---\nManual closing.' });
+  const acceptedCount = element({ textContent: '' });
+  const copy = element({ textContent: 'Copy', dataset: { labelSuccess: 'Copied' } });
+  const writes = [];
+  const findings = ['concise-output', 'duplicate-publishing-rule'].map((id) => {
+    const classes = new Set();
+    const label = { textContent: 'Accepter' };
+    const input = element({
+      checked: false,
+      dataset: { prompt: `Apply ${id}.`, labelOff: 'Accepter', labelOn: 'Accepté' },
+    });
+    const finding = {
+      dataset: { id },
+      classes,
+      input,
+      label,
+      classList: { toggle: (name, enabled) => enabled ? classes.add(name) : classes.delete(name) },
+      querySelector: (selector) => ({
+        '.accept-input': input,
+        '.accept-toggle span': label,
+      })[selector],
+    };
+    input.closest = () => finding;
+    return finding;
+  });
+  const document = {
+    querySelector: (selector) => ({
+      '#report': { dataset: { labelAcceptedOne: '{count} acceptée', labelAcceptedOther: '{count} acceptées' } },
+      '#execution-prompt': prompt,
+      '#accepted-count': acceptedCount,
+      '#copy-prompt': copy,
+    })[selector] || null,
+    querySelectorAll: (selector) => ({
+      '.finding': findings,
+      '.accept-input': findings.map(({ input }) => input),
+    })[selector] || [],
+  };
+
+  runInNewContext(readFileSync(resolve(
+    __dirname,
+    '../../plugins/aidd-refine/skills/05-improve/assets/report.js',
+  ), 'utf8'), {
+    document,
+    navigator: { clipboard: { writeText: async (value) => writes.push(value) } },
+    window: { setTimeout() {} },
+  });
+  return { prompt, acceptedCount, copy, findings, writes };
+}
+
+test('acceptance uses metadata IDs without visible ID nodes and retains manual edits', () => {
+  const { prompt, acceptedCount, findings } = loadReport();
+  const second = findings[1];
+  second.input.checked = true;
+  second.input.dispatch('change');
+
+  assert.ok(second.classes.has('is-accepted'));
+  assert.equal(second.label.textContent, 'Accepté');
+  assert.equal(acceptedCount.textContent, '1 acceptée');
+  assert.match(prompt.value, /\[duplicate-publishing-rule\] Apply duplicate-publishing-rule\./);
+
+  prompt.value = prompt.value.replace('Apply duplicate-publishing-rule.', 'Keep my edited instruction.');
+  second.input.checked = false;
+  second.input.dispatch('change');
+  assert.ok(!second.classes.has('is-accepted'));
+  assert.equal(second.label.textContent, 'Accepter');
+  assert.equal(acceptedCount.textContent, '0 acceptées');
+  assert.doesNotMatch(prompt.value, /\[duplicate-publishing-rule\]/);
+
+  second.input.checked = true;
+  second.input.dispatch('change');
+  assert.match(prompt.value, /\[duplicate-publishing-rule\] Keep my edited instruction\./);
+  assert.match(prompt.value, /^Manual introduction\./);
+  assert.match(prompt.value, /Manual closing\.$/);
+});
+
+test('copy reads the current editable prompt after acceptance', async () => {
+  const { prompt, copy, findings, writes } = loadReport();
+  findings[0].input.checked = true;
+  findings[0].input.dispatch('change');
+  prompt.value += '\nAn extra manual instruction.';
+
+  await copy.dispatch('click');
+
+  assert.deepEqual(writes, [prompt.value]);
+  assert.equal(copy.textContent, 'Copied');
+});