feat(review): add cached repo quality-culture profile as AI-review grounding

[JSONbored]

Summary

• Implements issue feat(review): infer a cached per-repo quality-culture profile as AI-review grounding context #2995: infer a cached per-repo quality-culture profile from the repo's own merge history and feed it into the AI reviewer's prompt as additive grounding context.
• New shared module src/review/repo-culture-profile.ts (extractRepoCultureProfile, deriveRepoCultureProfile) derives a deterministic profile — typical merged-PR size (median changed-file count, banded tiny/small/medium/large), typical PR description length, and common accepted labels — purely from recent_merged_pull_requests rows already stored by this engine. Below 5 sampled merged PRs it returns an explicit present: false insufficient-data branch, never a partial guess.
• Cached per repo in the existing signal_snapshots table (same mechanism signals/focus-manifest-loader.ts already uses for the manifest cache), invalidated on either a 6-hour TTL or a merged-PR-count drift check (a cheap COUNT(*), not a row re-scan) — matching the issue's "TTL or new commits" invalidation ask with the simplest signal already available.
• New host adapter src/review/repo-culture-profile-wire.ts (mirrors review/rag-wire.ts's structure exactly) formats the profile into a self-labelled, reference-only prompt block and is gated by a new GITTENSORY_REVIEW_CULTURE_PROFILE env flag plus a new per-repo .gittensory.yml review.culture_profile: true opt-in (parsed/resolved in signals/focus-manifest.ts the same way review.effort_score/review.changed_files_summary are). Both default off — the reviewer prompt is byte-identical until a repo deliberately opts in on top of the global flag being enabled.
• Wired into services/ai-review.ts (GittensoryAiReviewInput.cultureProfileContext, appended in buildUserPrompt right after the existing RAG section) and into queue/processors.ts's runAiReviewForAdvisory, following the exact same fetch/gate/pass-through pattern as the RAG feature. Explicitly ADVISORY GROUNDING ONLY — it never becomes a gate/scoring input and never changes the structured AI-review output contract.
• Bumped AI_REVIEW_CACHE_INPUT_VERSION (ai-review-input:v1 → v2) since the AI-review result cache's fingerprint features object gained a cultureProfile member; every field is now explicitly enumerated (not passed through raw) so a future new feature key can't silently invalidate every existing cache entry again.
• Confirmed via investigation (per the issue) that the existing src/review/repo-profile.ts (extractRepoProfile) is for a different epic (Roadmap: repo-specific CLAUDE.md/AGENT.md + skills generation & maintenance #2993/feat(review): extract a codebase-grounded repo profile from RAG + signals #2999, repo-doc/CLAUDE.md generation from the RAG code index) and does not carry PR-size/label/history signals — this PR does not repurpose it, it adds a new, narrowly-scoped module for the commit/PR-history signal this issue actually asks for.
• Design decision (no ambiguity to escalate, noted for transparency): PR "size" is derived from changedFiles.length (already reliably populated by the existing merged-PR backfill), not from additions/deletions in the untyped raw GitHub payload, since the latter isn't part of the typed GitHubPullRequestPayload shape and would be a less certain source of truth.

Scope

• The PR title follows type(scope): short summary Conventional Commit format.
• This PR is focused and does not mix unrelated backend, UI, MCP, docs, dependency, and deploy changes.
• This follows CONTRIBUTING.md and does not reintroduce GitHub Pages, VitePress, site/, or CNAME.
• Linked issue: closes feat(review): infer a cached per-repo quality-culture profile as AI-review grounding context #2995.

Validation

• git diff --check
• npm run actionlint
• npm run typecheck
• npm run test:coverage (targeted, on the touched/new files — see coverage numbers below); 100% statements/branches/functions/lines on the two new modules (repo-culture-profile.ts, repo-culture-profile-wire.ts).
• npm run test:workers
• npm run build:mcp
• npm run test:mcp-pack
• npm run ui:openapi:check
• npm run ui:lint (0 errors; only the pre-existing repo-wide react-refresh/only-export-components warnings, unrelated to this change)
• npm run ui:typecheck / npm run ui:build — skipped: both fail in this sandbox with Cannot find package '@lovable.dev/vite-tanstack-config', reproduced identically on a clean origin/main checkout with no changes applied (a private-registry package unavailable in this environment), so it is a pre-existing local-environment limitation, not something introduced by this PR.
• npm audit --audit-level=moderate (0 vulnerabilities)
• New/changed behavior has unit tests for both branches of every conditional touched (sparse-history fallback vs. populated profile; cache hit vs. miss vs. TTL-stale vs. count-drift-stale; malformed/sparse cached-payload reconstruction; fail-safe on a throwing storage read/write; flag-on/flag-off prompt parity).

Also ran, beyond the template list, since this touches generated config surfaces: npm run cf-typegen (new GITTENSORY_REVIEW_CULTURE_PROFILE wrangler var), npm run docs:drift-check, npm run db:schema-drift:check, npm run db:migrations:check, npm run ui:openapi:settings-parity — all clean, no drift.

Safety

• No secrets, wallet details, hotkeys, coldkeys, user PATs, private keys, raw trust scores, private rankings, or private maintainer evidence are exposed.
• Public GitHub text stays sanitized; the new prompt section is explicitly labelled reference-only/non-binding and is never surfaced publicly (it only ever reaches the AI reviewer's own prompt).
• [N/A] No auth, cookie, CORS, GitHub App, Cloudflare session changes.
• [N/A] No API/OpenAPI/MCP surface change (a .gittensory.yml-only + env-flag-only knob, same pattern as review.effort_score).
• [N/A] No UI changes beyond two doc pages (docs.tuning.tsx, docs.privacy-security.tsx) documenting the new env flag in prose, matching the existing flag-list format exactly — no visual/interactive change.
• Public docs updated: .gittensory.yml.example / config/examples/gittensory.full.yml (new review.culture_profile field documented), plus the two UI docs routes above.

Notes

• Rebased cleanly onto latest origin/main (including a real merge conflict against the concurrently-merged Boundary-safe test generation #1972 boundary-safe test-generation advisory PR, which added sibling fields to the same FocusManifestReviewConfig literals — resolved by combining both PRs' fields, verified with a full re-run of the affected test suites post-rebase).
• Kept the diff scoped to exactly this issue: no adjacent refactors, no touching the Autonomous Miner System's merge-bar inference (out of scope per the issue's own phrasing — "the two consumers never drift" only requires the extractor be importable by both, which it now is; wiring the miner side is left to that system's own follow-up).

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	gittensory-ui	94cbd15	Commit Preview URL Branch Preview URL	Jul 06 2026, 12:36 PM

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.40%. Comparing base (ee686e0) to head (94cbd15).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3802      +/-   ##
==========================================
+ Coverage   93.38%   93.40%   +0.01%     
==========================================
  Files         323      325       +2     
  Lines       32727    32823      +96     
  Branches    11982    12022      +40     
==========================================
+ Hits        30563    30659      +96     
  Misses       1530     1530              
  Partials      634      634              

Files with missing lines	Coverage Δ
src/queue/processors.ts	94.43% <100.00%> (+<0.01%)	⬆️
src/review/ai-review-cache-input.ts	100.00% <100.00%> (ø)
src/review/repo-culture-profile-wire.ts	100.00% <100.00%> (ø)
src/review/repo-culture-profile.ts	100.00% <100.00%> (ø)
src/services/ai-review.ts	97.01% <100.00%> (+0.01%)	⬆️
src/signals/focus-manifest.ts	99.12% <100.00%> (+<0.01%)	⬆️

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[JSONbored]

Fixed both real issues found by CI:

1. `cf-typegen` drift — regenerated `worker-configuration.d.ts` (stale against the wrangler/workers-types versions main has moved to since this branch was last rebased).
2. Superagent P3 — `entry.label` (user-controlled GitHub label text from merged PRs) flowed into the AI reviewer prompt with no sanitization. Fixed via the existing `neutralizePromptInjection` defense (same one `safeReviewTitle` already uses for untrusted PR titles).
3. `codecov/patch` (97.98%, 2 partial branches) — added the missing `review.culture_profile` round-trip test in `focus-manifest.test.ts` (every sibling boolean flag already has one), and a new webhook-level test covering `dynamicReviewFeatures.cultureProfile` — unlike its env-only siblings (grounding/rag/enrichment/reputation), this one is dual-gated by both the global env flag AND the per-repo manifest opt-in, so no existing test drove both true through a real review.

1207 tests pass across all touched/related files, typecheck/actionlint/audit clean.

[loopover-orb]

Important

🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪

🔍 Gittensory is reviewing…

AI analysis is in progress. This comment will update when the review is complete.

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed · 🟪 Reviewing}