feat(review): post quiet inline review comments on the PR diff - #1528
Merged
Conversation
PR B activates the CodeRabbit-style inline-comments feature (PR A #1525 shipped the inert contract). On top of the decision summary, the AI reviewer's line-anchored findings are posted as ONE non-blocking review (event: COMMENT, never REQUEST_CHANGES) so a contributor sees exactly what to fix for a resubmission without the gate ever changing. - createPullRequestReviewComments: a quiet COMMENT-event review with inline comments, threading the action mode (dry-run suppression). - inline-comments.ts: rightSideLinesFromPatch validates each finding's line against the diff hunks (out-of-diff lines dropped → never a 422), selectInlineComments dedupes/caps, postInlineReviewComments is fail-safe (swallows API errors, never touches the gate), and maybePostInlineComments posts once per head SHA — a cache hit carries no findings, so the ~2-min re-gate sweep never reposts, and the off-path does zero extra work. - Gated at three layers (all default OFF): GITTENSORY_REVIEW_INLINE_COMMENTS, the GITTENSORY_REVIEW_REPOS allowlist, and the per-repo review.inline_comments manifest toggle. - Docs + .env.example + flag plumbing (env.d.ts, wrangler.jsonc, cf-typegen).
Contributor
|
Superagent didn't find any vulnerabilities or security issues in this PR. |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #1528 +/- ##
==========================================
+ Coverage 95.46% 95.47% +0.01%
==========================================
Files 195 196 +1
Lines 21148 21197 +49
Branches 7649 7662 +13
==========================================
+ Hits 20188 20237 +49
Misses 383 383
Partials 577 577
🚀 New features to boost your workflow:
|
andriypolanski
pushed a commit
to andriypolanski/gittensory
that referenced
this pull request
Jun 27, 2026
…lapse nits (JSONbored#1094) Soak finding: the bot FALSE-CLOSED a good contributor PR (metagraphed JSONbored#1528, green CI, mergeable) on a hallucinated blocker — it claimed loadArtifactData was 'not imported -> ReferenceError', but that function is defined locally and used by ~24 call sites. Root cause: the model sees only the DIFF, not the whole file, so new code referencing an existing (out-of-diff) symbol looks undefined; both free models made the same mistake -> consensus -> close. (Full-file grounding is enabled but did not prevent it, so a prompt-level guard is the reliable fix.) - DIFF SCOPE guard in the review prompt: never report a missing import / undefined symbol / 'X not defined -> ReferenceError' as a blocker unless the diff ITSELF removes the definition or introduces the symbol without defining it; otherwise it is at most a nit. Directly kills the false-close-on-hallucination class. - Nits now render inside a collapsed <details> toggle (assessment + blockers stay visible), per request — keeps the comment focused.
andriypolanski
pushed a commit
to andriypolanski/gittensory
that referenced
this pull request
Jun 27, 2026
…rdening (JSONbored#1099) * fix(guardrail): fail-closed on KV outage + broaden crucial-path coverage Flood-readiness (emission weights rose on metagraphed + gittensory; a contributor wave is imminent). Two guardrail hardenings: 1. Fail-CLOSED on a KV read fault. loadHardGuardrailGlobs previously fell back to the narrow default (.github/workflows + scripts) on a THROWN read, so a KV outage correlated with a flood would silently shrink the guarded surface and let crown-jewel edits (scoring/auth/rules/the gate) auto-merge. A thrown read now returns FAIL_CLOSED_GUARDRAIL_GLOBS = ['**'] → every PR is held for human review until the read recovers. A legitimately-absent key still uses the narrow default so a freshly-installed repo can operate. 2. Broadened the LIVE per-repo KV globs (done out-of-band, effective immediately): - gittensory: added src/upstream/**, src/settings/**, src/review/**, src/services/**, src/github/**, src/config/** — the gate/decision/reviewer/ scoring/auth engine that lived OUTSIDE the old dir-prefix guards (e.g. score-breakdown.ts, command-authorization.ts, agent-actions.ts, ai-review.ts, ruleset.ts). This is the awesome-claude JSONbored#4196 incident class. - metagraphed: added src/** (the old config guarded specific .mjs files but NOT src/mcp-server.mjs — the file PR JSONbored#1528 touched). Tests: change-guardrail asserts the crucial out-of-dir files now hit the guardrail while infra/data/registry/docs/tests still auto-merge; guardrail-config asserts the outage path fails closed. Full suite green (3489). * fix(review): break the Gittensory Gate self-deadlock that froze reviews under load ROOT CAUSE of the stall (diagnosed from the LIVE audit_events ledger, not the frozen review_targets/review_audit): the Gittensory Gate is a REQUIRED check, and fetchLiveCiAggregate counted it like any other CI. A review posts the gate as in_progress, then prReadyForReview waits for ALL required CI to finish — but the gate never finishes (it's only concluded by the very review being deferred). So every green-CI PR deferred forever as 'CI still running' (38 review_deferred_ci_pending in 30 min; metagraphed JSONbored#1532 deferred 4x with all checks green). Fix: fetchLiveCiAggregate now SKIPS the bot's own GITTENSORY_GATE_CHECK_NAME in both the check-runs and commit-status loops, so the bot waits on real CI only and never on its own gate. Test: a green 'test' check + an in_progress 'Gittensory Gate' (gate in the required set) now reports ciState 'passed', not 'pending'.
andriypolanski
pushed a commit
to andriypolanski/gittensory
that referenced
this pull request
Jun 27, 2026
…g (#hold-crucial-on-reject) (JSONbored#1109) Operator decision: a guarded/crucial PR (CI, the review engine, visual) must NEVER be auto-closed, even on a reject verdict — a hallucinated reject (the JSONbored#1528 near-miss: diff-only false-positives) must not auto-close a good crucial PR. The owner verifies + closes/merges. The BULK (non-guarded) contributor PRs still auto-close one-shot on a bad verdict / conflict; only the small crucial set is held. Restores the guardrail check on willClose (reverts JSONbored#1106's removal). Owner/automation PRs never close regardless.
andriypolanski
pushed a commit
to andriypolanski/gittensory
that referenced
this pull request
Jun 27, 2026
…er + brace-aware JSON parse (#accuracy-gap-1, #accuracy-gap-3) (JSONbored#1110) Two reviewbot-parity fixes from the accuracy-gap audit, targeting the metagraphed JSONbored#1528 class (model hallucinates 'missing import / undefined symbol' on code it cannot see). GAP-1 (the JSONbored#1528 root): buildAiReviewDiff was a blind 60k head-slice that break-DROPPED whole files in stored order — no source-first ordering, no hunk-aware reduction. On a multi-file PR the file that DEFINES a symbol could be dropped while another referenced it, so the reviewer 'saw' an undefined symbol (survived even with full-file grounding on). Port reviewbot src/core/diff.ts into a new guarded src/review/review-diff.ts: order SOURCE first (lockfiles/generated/docs/tests drop first), reduce oversized patches hunk-aware (keep highest-signal hunks) instead of dropping, always list patch-less/ over-budget files with counts, and raise the budget 60k->80k (the 120B models are 128k-context). GAP-3: parseModelReview used a greedy /{[\s\S]*\}/ that spanned first-{ to last-} and swallowed the gpt-oss/nemotron <think> scratchpad object together with the verdict, silently corrupting/dropping reviews. Port reviewbot's brace-depth-aware, string-safe extractLastJsonObject (returns the LAST complete top-level object). Full suite green (3499). New: review-diff.test.ts (source survives over lockfiles under tight budget; patch-less file listed; hunk-aware reduction) + ai-review.test.ts scratchpad+fenced regression tests.
2 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
PR B of 2 — activates the CodeRabbit-style quiet inline-comments feature (PR A #1525 shipped the inert contract). On top of the decision summary, the AI reviewer's line-anchored findings post as one non-blocking review (
event: COMMENT, neverREQUEST_CHANGES) — a contributor sees exactly what to fix for a resubmission, and the gate is never affected.createPullRequestReviewComments— a quietCOMMENT-event review carrying inline comments, threading the actionmode(so a dry-run instance suppresses the write).inline-comments.ts—rightSideLinesFromPatchvalidates each finding's line against the diff hunks (out-of-diff lines are dropped → never a 422);selectInlineCommentsdedupes + caps;postInlineReviewCommentsis fail-safe (swallows API errors, never touches the gate);maybePostInlineCommentsposts once per head SHA — a cache hit carries no findings, so the ~2-min re-gate sweep never reposts, and the off-path does zero extra work (not even a PR-files read).GITTENSORY_REVIEW_INLINE_COMMENTS+ theGITTENSORY_REVIEW_REPOSallowlist + the per-repo.gittensory.ymlreview.inline_commentstoggle.Scope
src/review/inline-comments.ts(new),src/github/pr-actions.ts,src/queue/processors.tssrc/env.d.ts,wrangler.jsonc,worker-configuration.d.ts(cf-typegen)review-configuration.md,self-hosting.md,.env.exampleinline-comments(new),github-pr-actionsValidation
npm run typecheckclean ·npm run test:coverage4641 passed · 100% line+branch on the diff (inline-comments.ts, the new GitHub method, and the processor wiring all fully covered; remaining flags in the touched files are pre-existing) ·ui:openapi:checkcleanSafety
event: COMMENTis hardcoded (can't be a blocking review). Default-OFF at every layer; fully fail-safe (no inline-post failure can affect the gate).