v0.8.48: search improvements, file versions

[waleedlatif1]

• docs(blog): update enterprise (docs(blog): update enterprise #8002)
• improvement(knowledge): skip redundant document resurrection writes (improvement(knowledge): skip redundant document resurrection writes #8000)
• fix(file-search): skip base64-dominated files and redact query errors from task failures (fix(file-search): skip base64-dominated files and redact query errors from task failures #8004)
• feat(files): workspace file version history over the v2 API, CLI, and MCP (feat(files): workspace file version history over the v2 API, CLI, and MCP #7997)
• improvement(knowledge): remember a caller's saturated search reach (improvement(knowledge): remember a caller's saturated search reach #8006)
• fix(knowledge): keep Slack searchable while its member crawl runs (fix(knowledge): keep Slack searchable while its member crawl runs #8005)
• improvement(knowledge): rank organization keyword search with Tin (improvement(knowledge): rank organization keyword search with Tin #8007)
• fix(files): file version history follow-ups (fix(files): file version history follow-ups #8010)
• feat(google-calendar): add RSVP operation to respond to invitations (feat(google-calendar): add RSVP operation to respond to invitations #8019)
• fix(chat): return 403 for inactive chats without writing an execution log (fix(chat): return 403 for inactive chats without writing an execution log #8024)
• fix(api): refuse workspace keys and enforce personal_api_key.use on v1 audit logs (fix(api): refuse workspace keys and enforce personal_api_key.use on v1 audit logs #8025)
• fix(sso): require a session to list SSO providers (fix(sso): require a session to list SSO providers #8023)
• fix(self-hosting): stop publishing Postgres and require POSTGRES_PASSWORD (fix(self-hosting): stop publishing Postgres and require POSTGRES_PASSWORD #8035)
• fix(realtime): validate cursor and selection presence payloads (fix(realtime): validate cursor and selection presence payloads #8041)
• fix(copilot): stop returning generate_api_key plaintext on the in-band tool route (fix(copilot): stop returning generate_api_key plaintext on the in-band tool route #8042)
• improvement(copilot): refuse approval-gated tools on the in-band lane (improvement(copilot): refuse approval-gated tools on the in-band lane #8044)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Sep 20, 2026 2:11am UTC

[greptile-apps]

Confidence Score: 1/5

The PR is not safe to merge until Slack reply deletions stop remaining searchable for weeks and database-query redaction removes the raw parameter-bearing cause from Trigger task failures.

Findings

1.   Slack replies remain searchable ▶
2.   Redaction preserves sensitive cause ▶
3.  Indented base64 bypasses detection ▶

Summary

This PR adds workspace-file version history across storage, v2 APIs, CLI/MCP, retention, and audit surfaces; revises Slack synchronization and member-access renewal; introduces Tin-backed organization keyword ranking; and hardens workspace-file search indexing.

• Adds version creation, listing, reading, downloading, reverting, deleting, and retention cleanup.
• Changes Slack thread hashing and scheduled hydration to reduce redundant source reads.
• Adds Tin keyword projections, migration/backfill support, and search fallback logic.
• Excludes base64-dominated files and redacts database failures from indexing tasks.
• Includes substantial API, migration, integration-test, documentation, and generated-client updates.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Workspace file write] --> B[Version transaction]
  B --> C[(workspace_file_version)]
  B --> D[Current file object]
  C --> E[v2 API / CLI / MCP]
  C --> F[Retention cleanup]
  F --> G[Storage cleanup outbox]

  H[Slack listing] --> I{Root changed, active,<br/>or rolling refresh?}
  I -->|Yes| J[Hydrate thread]
  I -->|No| K[Reuse indexed document]
  J --> L[Knowledge index]
  K --> L

  M[Organization keyword query] --> N{Tin ready?}
  N -->|Yes| O[Tin candidate ranking]
  N -->|No| P[GIN fallback]
  O --> Q[Access filtering and hydration]
  P --> Q

Loading

_{Reviews (1) · Last reviewed commit: "improvement(knowledge): rank organizatio..."}

[greptile-apps]

Slack replies remain searchable

For quiet threads, this code skips hydration whenever the root-derived version is unchanged. That version includes only the root edit timestamp, reply count, and latest-reply timestamp, so editing or deleting an existing reply does not change it. Once the seven-day active window passes, stale or deleted reply text can remain indexed and searchable until the thread's one-in-28-day refresh.

How this was verified: The listing hash excludes reply edit and deletion state, and a matching root version is classified as current without hydrating the thread.

Knowledge Base Used:

• Knowledge platform
• Knowledge ingestion, access, and search

[greptile-apps]

Redaction preserves sensitive cause

The sanitized error still attaches the original Drizzle error as its cause, even though that error carries the SQL and bound file text this helper is meant to redact. The indexing task throws this wrapper to Trigger.dev, so cause-chain serialization can retain confidential file content. Equivalent background database-error paths avoid this by throwing a cause-free diagnostic.

How this was verified: The wrapper directly retains the original parameter-bearing Drizzle error as its cause before the indexing task throws it.

[greptile-apps]

Indented base64 bypasses detection

Wrapped base64 is joined across lines only when the run starts at column zero. Common indented 64- or 76-column payloads in YAML, generated source, and nested text are therefore split into runs shorter than 256 characters and report no encoded bytes. These base64-dominated files proceed into the expensive trigram index instead of being skipped. Please track wrapped runs independently of leading indentation and add coverage for an indented payload.

[cubic-dev-ai]

4 issues found across 137 files

Confidence score: 2/5

• apps/sim/lib/core/errors/database-query-error.ts can expose SQL and bound parameters through the DrizzleQueryError cause chain, allowing indexed file content to bypass redaction; omit the original query error as cause.
• apps/sim/lib/knowledge/search/tin-keyword.ts may retain a true cache value for ten minutes after Tin rows are deleted, causing selectTinPage to treat empty pages as successful and return no matches; invalidate or refresh this state when an index is demoted.
• apps/sim/connectors/slack/slack.ts can treat a quiet thread as current from rootVersion even after an existing reply is edited or deleted, returning stale thread data; track reply edit/deletion state or force hydration before accepting the listing.
• apps/sim/lib/workspace-files/search/index-plan.ts splits indented wrapped payloads at every newline when runFillsLines is false, so base64-heavy files may evade exclusion; track payload line width independently of indentation.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/sim/lib/knowledge/search/tin-keyword.ts">

<violation number="1" location="apps/sim/lib/knowledge/search/tin-keyword.ts:39">
P1: When a search index is demoted, this cache can keep `true` for ten minutes after the migration trigger deletes its Tin rows. `selectTinPage` treats `ranked = 0` as a successful page, so searches return no matches instead of falling back to GIN. Invalidate membership on `isSearchIndex` changes or avoid caching positive entries.</violation>
</file>

<file name="apps/sim/lib/core/errors/database-query-error.ts">

<violation number="1" location="apps/sim/lib/core/errors/database-query-error.ts:23">
P1: Do not attach the original query error as `cause` here. `DrizzleQueryError` retains SQL and bound parameters, so the indexing task can still expose file content through the supposedly redacted cause chain.</violation>
</file>

<file name="apps/sim/lib/workspace-files/search/index-plan.ts">

<violation number="1" location="apps/sim/lib/workspace-files/search/index-plan.ts:46">
P2: Indented wrapped payloads are split at every newline because `runFillsLines` is false after leading whitespace. Track payload line width independently of indentation so base64-dominated files are excluded from the expensive trigram index.</violation>
</file>

<file name="apps/sim/connectors/slack/slack.ts">

<violation number="1" location="apps/sim/connectors/slack/slack.ts:545">
P1: A quiet thread is marked current from `rootVersion` alone, but Slack does not change that value when an existing reply is edited or deleted. Track reply edit/deletion state or force hydration before accepting this listed hash, otherwise stale reply text remains searchable.</violation>
</file>

<sub>Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic</sub>

[cubic-dev-ai]

P1: When a search index is demoted, this cache can keep true for ten minutes after the migration trigger deletes its Tin rows. selectTinPage treats ranked = 0 as a successful page, so searches return no matches instead of falling back to GIN. Invalidate membership on isSearchIndex changes or avoid caching positive entries.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/sim/lib/knowledge/search/tin-keyword.ts, line 39:

<comment>When a search index is demoted, this cache can keep `true` for ten minutes after the migration trigger deletes its Tin rows. `selectTinPage` treats `ranked = 0` as a successful page, so searches return no matches instead of falling back to GIN. Invalidate membership on `isSearchIndex` changes or avoid caching positive entries.</comment>

<file context>
@@ -0,0 +1,95 @@
+/** Only organization search indexes are projected; `is_search_index` is fixed at creation. */
+const searchIndexBases = new LRUCache<string, boolean>({
+  max: 10_000,
+  ttl: SEARCH_INDEX_TTL_MS,
+})
+
</file context>

[cubic-dev-ai]

P1: Do not attach the original query error as cause here. DrizzleQueryError retains SQL and bound parameters, so the indexing task can still expose file content through the supposedly redacted cause chain.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/sim/lib/core/errors/database-query-error.ts, line 23:

<comment>Do not attach the original query error as `cause` here. `DrizzleQueryError` retains SQL and bound parameters, so the indexing task can still expose file content through the supposedly redacted cause chain.</comment>

<file context>
@@ -8,3 +8,18 @@ import { DrizzleQueryError } from 'drizzle-orm/errors'
+  const code = getPostgresErrorCode(queryError) ?? 'no error code'
+  const reason = getPostgresCancellationReason(queryError)
+  return new Error(`${operation} failed (${reason ? `${code}, ${reason}` : code})`, {
+    cause: error,
+  })
+}
</file context>

[cubic-dev-ai]

P1: A quiet thread is marked current from rootVersion alone, but Slack does not change that value when an existing reply is edited or deleted. Track reply edit/deletion state or force hydration before accepting this listed hash, otherwise stale reply text remains searchable.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/sim/connectors/slack/slack.ts, line 545:

<comment>A quiet thread is marked current from `rootVersion` alone, but Slack does not change that value when an existing reply is edited or deleted. Track reply edit/deletion state or force hydration before accepting this listed hash, otherwise stale reply text remains searchable.</comment>

<file context>
@@ -403,7 +448,142 @@ function listingToken(syncContext?: Record<string, unknown>): string {
+  const next = parseContentHash(candidate)
+  const previous = parseContentHash(stored)
+  if (!next || previous?.kind !== 'hydrated') return 'stale'
+  if (next.kind === 'listed') return previous.version === next.version ? 'current' : 'stale'
+  return previous.text === next.text ? 'equivalent' : 'stale'
+}
</file context>

[cubic-dev-ai]

P2: Indented wrapped payloads are split at every newline because runFillsLines is false after leading whitespace. Track payload line width independently of indentation so base64-dominated files are excluded from the expensive trigram index.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At apps/sim/lib/workspace-files/search/index-plan.ts, line 46:

<comment>Indented wrapped payloads are split at every newline because `runFillsLines` is false after leading whitespace. Track payload line width independently of indentation so base64-dominated files are excluded from the expensive trigram index.</comment>

<file context>
@@ -2,11 +2,79 @@ import { Buffer } from 'node:buffer'
+    const isLower = code >= 97 && code <= 122
+    const isDigit = code >= 48 && code <= 57
+    if (isUpper || isLower || isDigit || code === 43 || code === 47) {
+      if (run === 0) runFillsLines = lineChars === 0
+      run++
+      lineChars++
</file context>