Phase B — verification instead of reading#13
Merged
Conversation
- CiSummary moves to model (domain rollup); AgentMode::Review (advisory pre-pass, never advances the gate); FindingSeverity/ReviewFinding; ConversationKind/ConversationItem (read-only GitHub context, distinct from ephemeral Comments per §0.4) - Review gains ci_summary, review_findings, conversation, last_reviewed_sha (all serde-default; legacy state loads clean) - AgentPrompts.review override; Config.notify_poll_secs; REVIEW_INSTRUCTION - bindings regenerated; FE exhaustive switches extended Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…se F Verified competitor teardowns (Graphite taxonomy/downvote/rules model, CodeRabbit price band + review/rework collapse, the ~58% critical-bug ceiling), HCI evidence for the gated loop (four oversight surfaces, 2.1-steps-per-prompt chunking, zero-of-99 autonomy trust), the trust-miscalibration hazard binding our evidence/trajectory UI, and Cisco sizing mechanics. Locks 10 design decisions; adds roadmap Phase F. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
git2 blob reads at a revision (binary/oversize/missing-path -> None, bad rev -> Err) and a gh api contents fallback with pure, tested response parsing. 512 KiB cap guards Monaco full-file rendering. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Tolerant JSON-array extraction (string-aware balanced scan, junk-wrapped agent output survives), skip-malformed entries, stable content-hashed ids, 200-finding cap. assemble_review_prompt mirrors the plan prompt sections with a golden; findings_file_path shares the plan slug sanitizer. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…, weakening Pure single-pass unified-diff analysis: size classes at documented boundaries, priority-ordered per-file risk flags, assertion counting, and seven test-weakening detectors with jump targets. 16 tests + regenerated bindings. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- get_evidence: deterministic signals + CI rollup (trajectory slot ready for Phase D) - pre_review: explicit reviewer-agent spawn (AgentMode::Review); findings ingested on completion, file deleted after transport, gate state never touched; findings cleared when the head advances - get_file_pair: worktree git2 reads with gh contents fallback and honest full:false degradation Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…le view EvidenceStrip (CI/size/test-delta/risk/weakening chips with jump-to-hunk), advisory finding pins on parallel zone machinery (dashed severity rail, local dismiss, never counted as comments), Pre-review button, full-file toggle with identity line maps + per-head cache, TS signal mirror with Rust-fixture parity tests. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ails Stale pins on changed content mislead; the head has moved regardless of whether the gh re-fetch succeeded (reviewer minor #6). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
LifeLex
added a commit
that referenced
this pull request
Jul 2, 2026
alejandro-nexcade
added a commit
that referenced
this pull request
Jul 2, 2026
* core+app: Phase E backend — conversation ingest, teammate interdiff, Linear intent fetch_conversation merges reviews/review-comments/issue-comments (three concurrent gh reads, fixture-tested parsers, state-only approvals kept); submit_github_review records last_reviewed_sha; get_teammate_interdiff compares it to the pinned head; Linear descriptions seed review intent and capped plan-issue gists (goldens byte-identical). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * app(fe): Phase E — conversation band, teammate re-review on push Read-only GitHub conversation context for review-requested PRs (state- aware labels, plain-text bodies, external links); the interdiff toggle gains a teammate source driven by last_reviewed_sha vs pinned head with a 'pushed since your review' header note. Authored (dispatch_snapshot) and teammate paths are mutually exclusive by construction. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * app: bound Linear API calls with a 15s timeout A stalled Linear response was the one un-timed await in plan generation and kickoff; errors already degrade gracefully, now hangs do too. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * docs: mark roadmap phases B-E implemented (PRs #13-#16) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Alejandro Perez <alejandro.perez.gancedo@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
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
2 participants
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.
What (ROADMAP Phase B; realizes SPEC §8 step 1's ci_delta/test_count_delta)
diff_signalscore engine.#[ignore],|| true,.skip/.only, deleted test fns/files, snapshot rewrites) as danger chips with jump-to-hunk.AgentMode::Review): explicit Pre-review button spawns a reviewer agent whose JSON findings render as dismissible severity pins on parallel zone machinery. Never touches the gate, never counts as comments, cleared when the head advances (even if the diff re-fetch fails). Tolerant findings parser (junk-wrapped agent output survives), review prompt with golden.gh api contentsfallback, identity line maps, per-head cache, honestfull:falsedegradation.RESEARCH_INTERACTION_PATTERNS.md) + ROADMAP Phase F.Research constraints honored
Findings are a triage layer, never a verdict (AI reviewers cap at ~58% critical-bug catch even in vendor benchmarks); evidence strip is deterministic only — no model confidence displays (trust-miscalibration hazard, arXiv:2602.16844).
Verified
cargo fmt/clippy -D warnings/test --all (427+ green) · tsc strict · vitest 149 · vite build · reviewer-agent pass: ship-ready, zero blocking (6 minors: 1 fixed here, base-SHA pinning folds into Phase C's enrich unit, rest documented).
🤖 Generated with Claude Code