feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal by jinhongkuan · Pull Request #174 · BicameralAI/bicameral-mcp

jinhongkuan · 2026-05-04T01:02:26Z

Summary

_region_anchored_preflight now does a 1-hop code-graph expansion over caller-supplied file_paths before the binds_to lookup, so a decision bound to app/src/lib/git/reorder.ts surfaces when the caller passes the structurally-near app/src/ui/multi-commit-operation/reorder.tsx.
Direct-pin matches keep confidence=0.9; expansion-only matches get confidence=0.7 so callers can de-prioritize the latter without losing recall.
sources_chained includes "region_graph_expanded" when expansion contributed at least one new decision.

Linked issues

Closes #173.

Refs #168 (PostToolUse hook on bicameral_preflight) and #172 (caller-side discovery-first reminder) — this PR closes the recall ceiling those two PRs surfaced. Together they reconnect the contradiction-capture loop end-to-end.

Plan / Audit / Seal

Plan: trivial; risk:L1.

The change is additive: no MCP tool contract change (caller sends the same file_paths shape), no schema migration, falls back gracefully (returns input unchanged) when the symbol index is unavailable. Hub-file explosion is bounded by CodeLocatorConfig.max_neighbors_per_result per-symbol AND globally. Provenance is preserved on the response so downstream skills can decide whether to weight expanded matches differently.

Test plan

ruff check . — clean (220 files).
ruff format --check . — clean.
pytest tests/test_preflight_graph_expansion.py -v — 8/8 PASS:
- 6 adapter unit tests against an in-memory SymbolDB (1-hop neighbor walk, no-edge case, empty input, unindexed file, hub-file cap, uninitialized fallback).
- 2 handler integration tests against a real ledger with a fake code_graph stub (expansion adds path → decision surfaces + region_graph_expanded tag set; direct pin alone does NOT get tagged as expanded).
pytest tests/test_alpha_contract.py::test_preflight_surfaces_bound_decisions — direct-pin path still PASSes; the new tuple unpacking in _region_anchored_preflight doesn't break the existing contract.
pytest tests/test_preflight_hook.py tests/test_post_preflight_capture_hook.py tests/test_post_commit_sync_hook.py — 26/26 across the surrounding hook suites.
CI: e2e assertions (auto) — Flow 2a expected to flip ⚠️ FAIL → PASS. The chain that's been broken on every prior run is now expected to engage end-to-end:
1. Agent does Read/Grep/Glob discovery (fix(skill): preflight reminder allows discovery first, gates only writes #172).
2. Agent calls bicameral.preflight(topic, file_paths=[...]) — UI files OK.
3. Server expands by 1 hop, surfaces the bound decision (this PR).
4. PostToolUse hook (fix(skill): force resolve_collision via PostToolUse hook on preflight (#154) #168) fires its gate, injects Step 5.6 reminder.
5. Agent calls ingest(source=agent_session) + resolve_collision.
CI: ruff + mypy, MCP Regression Suite (ubuntu + windows), Schema Persistence Smoke, TruffleHog, Preflight Failure-Mode Eval (advisory) — all expected to pass.

Out of scope

Multi-hop traversal (>1) — defer until 1-hop telemetry shows recall is still inadequate.
Server-side topic→file resolution (BM25 / embeddings over decision descriptions). Different approach to the same recall problem; not coupled to the structural-graph approach.
MCP tool surface changes — bicameral.preflight request/response shape is byte-compatible; only confidence values and sources_chained set membership shift.

Carry-over plan

Lands on dev first. Will cherry-pick to triage-from-dev per the §10.5 lane convention so PR #165 picks up the fix before it merges to main.

…raversal (#173) ## Why The headline product loop — *dev prompt contradicts a prior decision → bicameral surfaces it → refinement captured → PM ratifies* — silently breaks when the caller-LLM picks ``file_paths`` that don't exactly match the file the prior decision is anchored to. Observed on the e2e runs for #168 and #172: agent passes UI-layer files (``reorder.tsx``, ``commit-list.tsx``) for the "reorder feature"; the seeded decision is bound to the git-layer file (``reorder.ts``); region-anchored lookup ``WHERE file_path IN $fps`` returns zero hits; preflight reports ``fired=false, decisions=[]``; the PostToolUse hook (#154 / #168) doesn't fire its gate (``fired && decisions[]``); Step 5.6 reminder never reaches the model; no ``ingest(agent_session)``, no ``resolve_collision``. Flow 2a fails advisory but the deeper consequence is real-world contradictions go uncaptured. The server has full structural knowledge of the codebase — the ``code_locator`` package ships a ``symbols`` + ``edges`` graph with ``invokes`` / ``imports`` / ``inherits`` / ``contains`` edges, exposed via ``SymbolDB.get_ego_graph(symbol_id, hops=1)``. Until now, ``_region_anchored_preflight`` did exact-match-only lookup and ignored the graph. This PR closes the gap. ## What - ``adapters/code_locator.py::RealCodeLocatorAdapter.expand_file_paths_via_graph`` — new public method. Walks each caller-supplied file's symbols, fetches each symbol's 1-hop ego graph, collects neighbor file paths. Returns ``(expanded, added)`` so callers can mark provenance. Bounded per-symbol AND globally by ``CodeLocatorConfig.max_neighbors_per_result`` to prevent hub-file explosion. Falls back gracefully (returns input unchanged) when the symbol index is unavailable. - ``handlers/preflight.py::_region_anchored_preflight`` — calls the expander before the ``binds_to`` lookup. Decisions reached via caller-supplied paths get ``confidence=0.9``; decisions reached only via expansion get ``confidence=0.7`` so the caller can de-prioritize the latter without losing recall. Returns ``(matches, used_graph_expansion)`` so the caller can record ``"region_graph_expanded"`` in ``sources_chained`` when expansion contributed at least one hit beyond the direct-match set. - ``skills/bicameral-preflight/SKILL.md`` Step 2 — documents the new expansion behaviour and the ``confidence`` / ``sources_chained`` semantics so downstream skills can read them. - ``tests/test_preflight_graph_expansion.py`` — 8 tests: - 6 adapter unit tests (synthetic ``SymbolDB``): finds 1-hop neighbor, handles no-edge / empty / unindexed inputs, caps hub-file explosion, falls back when uninitialized. - 2 handler integration tests (real ledger, fake ``code_graph``): expansion-only surfaces the bound decision and tags ``region_graph_expanded`` in ``sources_chained``; direct pin alone does not get tagged as expanded. ## Plan / Audit / Seal Plan: trivial; risk:L1 (additive — no contract change to the MCP tool surface, no schema migration, falls back gracefully when index is absent). Implementation note in commit body above. ## Test plan - [x] ``ruff check . && ruff format --check .`` — clean. - [x] ``pytest tests/test_preflight_graph_expansion.py -v`` — 8/8. - [x] ``pytest tests/test_alpha_contract.py::test_preflight_surfaces_bound_decisions`` — direct-pin path still PASSes (no regression on the existing contract). - [x] ``pytest tests/test_preflight_hook.py tests/test_post_preflight_capture_hook.py tests/test_post_commit_sync_hook.py`` — 26/26 across the related hook suites. - [ ] CI: ``e2e assertions (auto)`` — Flow 2a expected to flip ⚠️ FAIL → PASS now that preflight surfaces the bound decision under structural distance, the PostToolUse hook fires, and the agent has decision_ids to feed into ``ingest(agent_session)`` + ``resolve_collision``. Closes #173.

coderabbitai · 2026-05-04T01:02:35Z

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: bf43c57c-a5cb-4c0c-8ed0-d59ec9c5ff35

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch feat/preflight-graph-expansion

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

@silongtan

…ly, M6 eval Per @silongtan's review on #173 (cross-ref to the now-superseded #64 "Preflight: graph expansion for transitive relevance (M6)"), align PR #174 with that issue's design specifics: - **Edge-type filter**: narrow expansion to ``imports`` edges only. ``invokes`` / ``inherits`` / ``contains`` are symbol-level edges that over-broaden the file-level expansion contract. Imports is the file-level structural-dependency edge — matches the granularity of the region-anchored decision lookup. - **sources_chained tag**: rename ``"region_graph_expanded"`` → ``"graph"`` per #64. Sits cleanly alongside the existing ``"region"`` source tag. - **Input-seed cap**: ≤10 input seeds × ``max_neighbors_per_result`` neighbors per seed (#64's worst-case envelope). Prevents per-symbol walks from compounding when the caller passes a long ``file_paths`` list. - **M6 eval flips XFAIL → live**: ``tests/eval/preflight_dataset.jsonl`` M6 row now uses two new setup fields — ``region_decisions_pinned_to`` (path-aware decision lookup) and ``graph_neighbors`` (stub ``code_graph`` topology) — added to ``_apply_setup`` in ``run_preflight_eval.py``. The asserter now tests true graph-expansion semantics (mock returns the decision only when the expanded file_path is in the lookup set). - **Catalog flip**: ``docs/preflight-failure-scenarios.md`` M6 status goes ❌ → ✅ with a pointer to #173/#174. - **New regression test**: ``test_expander_filters_to_imports_only`` asserts that a neighbor reachable only via ``invokes`` / ``inherits`` does NOT appear in the expansion, locking the imports-only posture in. Test plan: - ``ruff check . && ruff format --check .`` — clean. - ``pytest tests/test_preflight_graph_expansion.py -v`` — 9/9 (added ``test_expander_filters_to_imports_only``). - ``pytest tests/eval/run_preflight_eval.py -v`` — 6 PASS + 3 XFAIL (M7 family, separate concern). M6 specifically flipped XFAIL → PASS. Deferred to follow-up (called out in the issue comment on #173): - ``preflight`` p95 baseline + 5ms perf assertion — needs the existing baseline harness to land first. - Splitting the implementation into a dedicated ``code_locator/tools/get_file_neighbors.py`` primitive vs the current method-on-adapter shape. Less load-bearing; revisit if a second consumer (e.g. drift detection) needs the same expansion.

…ming, no file paths Two changes that make Flow 2 a real test of the contradiction-capture loop instead of a partially-rigged one: 1. **Drop the "X is out, Y instead" framing.** The prior prompt told the agent the drag-drop approach was being replaced ("drag-and-drop is out", "no more drag-drop on this surface"). That short-circuits the contradiction-detection step the loop is supposed to validate — Step 5.6 fires when *the agent* notices the prompt conflicts with a surfaced decision, not when the user pre-announces the conflict. New prompt asks for a programmatic API for reordering commits with no replacement framing at all. The agent has to: a. read the surfaced "Drag-to-reorder commits, drag-to-squash, ..." decision and notice the prior plan was a drag-drop UX. b. read its own request (programmatic API, ordered SHA list, callable from any UI surface) and notice that's structurally incompatible with drag-drop. c. classify the situation as a supersede / refinement and execute Step 5.6. 2. **No file paths in the prompt.** The prior prompt named ``app/src/lib/git/reorder.ts`` directly. With #172's discovery- first contract, the caller LLM is supposed to use Grep/Glob/Read to identify files from a feature description on its own. Naming the file in the prompt cheats the discovery step — it's the agent's job to map "reordering commits" to ``reorder.ts``. Now the prompt describes the feature only; discovery is genuinely on the agent. If Flow 2a flips to PASS on this prompt, the end-to-end loop works on a clean test of the product claim: dev asks for a feature → agent discovers files → preflight surfaces the conflicting prior decision → agent recognizes the conflict → captures the refinement. If it stays advisory FAIL, that's an honest signal about which link is weakest (usually the agent's contradiction-judgment pass). Companion to PR #174 (#173 — graph expansion); kept on the same branch because both pieces are part of validating the same loop end-to-end.

…175) Even after #154 / #168 / #172 / #173, the contradiction-capture loop fails on prompts where the conflict is *semantic* rather than *lexical*. Latest PR #174 e2e: - Test prompt: "Add a programmatic API for reordering commits — takes an ordered list of commit SHAs and rewrites history to match. Wire it so any UI surface can call it..." - Server-side chain works end-to-end. Agent discovers files via Read/ Grep/Glob, calls preflight with file_paths populated, server surfaces the bound "Drag-to-reorder commits, drag-to-squash, ..." decision, PostToolUse hook fires its gate, Step 5.6 reminder reaches the model. - Agent's bicameral sequence: ['bicameral_preflight', 'bicameral_resolve_compliance']. No ingest(agent_session). No resolve_collision. The agent reads the surfaced decision, reads the user's prompt, reasons "adding an API doesn't necessarily replace a UI — they could coexist", and skips capture. That's a defensible engineering inference; LLMs aren't reliably capturing structural-mismatch refinements. Per #175's design discussion, route the judgment to the user instead of the agent. Changes: - skills/bicameral-preflight/SKILL.md — Step 5.6 rewritten. Step 5.6.1 is the new AskUserQuestion disambiguation (supersede / keep_both / unrelated); Step 5.6.2 is the mechanical capture, gated on the user picking one of the "yes" options. The new step fires whenever fired=True && len(decisions) > 0, regardless of guided mode (capture is the headline product behavior, not opt-in). - scripts/hooks/post_preflight_capture_reminder.py — reminder text retemplated. Instead of "you MUST call ingest+resolve_collision now", the reminder embeds the AskUserQuestion shape with the surfaced decision_id + description filled in, plus the answer→action branch table. Docstring rewritten to capture the new contract + the archaeology of why we abandoned the unconditional approach. - tests/e2e/run_e2e_flows.py::assert_flow_2a — pass criterion changes from "ingest+resolve_collision fired" to "AskUserQuestion was invoked with disambiguation shape after preflight surfaced decisions." The user-side response can't be driven in headless claude -p, so the testable signal is the question invocation itself. Loose shape match on options labels (supersede / keep_both / unrelated / "refinement of" / decision: ID). - tests/test_post_preflight_capture_hook.py — assertions updated to lock in the new D-path posture. Test renamed: test_reminder_is_unconditional_not_judgment_gated → test_reminder_routes_judgment_to_user_not_agent. Negative assertions catch regression to either the prior unconditional shape ("you MUST capture") or the original conditional shape ("IF you contradict ..."). - CHANGELOG.md — Changed section entries for the skill rewrite and the asserter contract change. Trust contract preserved: hook still fires only when fired=True && len(decisions) > 0. Silent on no signal. The disambiguation question runs at a moment the flow is already paused (rendering the surfaced block). Headless-mode caveat: AskUserQuestion in claude -p has no human to answer, so the e2e asserter checks invocation only — Step 5.6.2's mechanical capture only fires after a real human answer in interactive sessions. Acceptable per #175's acceptance criteria. Closes #175.

feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal

Bumps pyproject + RECOMMENDED_VERSION to 0.13.7 and resolves the stale git conflict markers that were committed into CHANGELOG.md by the previous `Merge branch 'main' into triage-from-dev` (c7d1274). v0.13.6 was bumped in pyproject on 2026-04-30 but never tagged or published to PyPI (latest published is v0.13.5; latest GitHub release is v0.13.5). v0.13.7 is the first release that ships everything merged into main since v0.13.5, including: - Preflight graph expansion + region anchored preflight (BicameralAI#173, BicameralAI#174) - Contradiction-capture flow via AskUserQuestion (BicameralAI#154, BicameralAI#175) - Preflight skill auto-fire fix on natural refactor prompts (BicameralAI#146) - SessionEnd hook re-entrancy + --auto-ingest (BicameralAI#147) - Post-preflight capture reminder hook (BicameralAI#168) - Flow1 asserter relax + flow2/2a split (BicameralAI#171) - v0 user flow e2e + demo recording carried over from dev (BicameralAI#165) - Lint-and-typecheck CI wired up; ruff format + fixes across 115 files See CHANGELOG.md for full details.

…Piece A) PR BicameralAI#174 closed the recall ceiling but introduced two silent fallback paths in `_region_anchored_preflight`: when `ctx.code_graph` was absent OR when the expander raised, the response shape was byte- identical to "expansion ran and matched zero" — caller couldn't tell recall was degraded. Three additive signals now surface every fallback (per Phase 2 spec posted on BicameralAI#243, all four open questions defaulted to recommended): 1. Response field — `sources_chained` includes `"graph_unavailable"`. Additive (never replaces existing `"region"` / `"graph"` tags). Bare tag — granular reason flows through telemetry, not the response shape, per signoff Q2. 2. Log level — exception case bumped from `logger.debug` → `logger.warning` with stable `[preflight:fallback]` substring + exception type for grep-friendly production logs. 3. Telemetry counter — new `preflight_telemetry.write_fallback_event( reason, session_id)` modeled on `write_ingest_refusal_event` (BicameralAI#216). Emits a `graph_expansion_fallback` row to the existing `~/.bicameral/preflight_events.jsonl` substrate. Reasons are a controlled enum: `"absent"`, `"missing_method"`, `"exception:<type>"`. Gated on `BICAMERAL_TELEMETRY=preflight`. The fallback case classifier in `_region_anchored_preflight` distinguishes three reasons (was conflated into a single `if expander is not None:` skip in the pre-BicameralAI#243 code): - `code_graph is None` → "absent" - `code_graph` set but no `expand_file_paths_via_graph` → "missing_method" - expander raised → "exception:<typ>" Skill update (`skills/bicameral-preflight/SKILL.md`) renders a one- line recall-degraded note to the agent when the tag is present: > Note: structural-neighbor lookup was unavailable this call — > recall may be reduced until the symbol index is rebuilt. Decisions > bound to files that import these may not have surfaced. Treats `"graph_unavailable"` as advisory: doesn't block the preflight surface; direct-pin matches are unaffected. Tests ----- 4 new cases in `tests/test_preflight_graph_expansion.py`: - test_preflight_fallback_absent_code_graph_tags_graph_unavailable — ctx with code_graph=None → response carries the tag, telemetry counter reason="absent" - test_preflight_fallback_expander_raises_warns_and_tags — stub expander raises RuntimeError → response carries the tag, `caplog` captures WARN-level log with `[preflight:fallback]` substring, telemetry counter reason="exception:RuntimeError" - test_preflight_successful_expansion_does_not_tag_graph_unavailable — regression guard: clean expansion path must NOT carry the tag (no false alarms) - test_preflight_empty_file_paths_does_not_tag_graph_unavailable — empty file_paths short-circuits before expansion check; the "expansion was never attempted" case is distinguishable from "attempted-and-fell-back" Existing tests use containment assertions (`"region" in sources_chained`) not exact list equality, so additive `"graph_ unavailable"` doesn't break them. What's NOT in this PR --------------------- Piece B (eager symbol-index initialization at server startup) is the follow-up commit on this branch. Lands separately so the response- shape change can ship without the adapter-lifecycle change. After both pieces land, the telemetry counter shipped here gives ongoing visibility into how often fallback engages in production. Refs BicameralAI#243 (parent BicameralAI#173 / PR BicameralAI#174). Plan signoff via BicameralAI#243 (comment). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…eralAI#243 Piece B) Pre-fix, the code-locator adapter had two cooperating problems that made silent fallback the default: 1. `get_code_locator()` returned a FRESH `RealCodeLocatorAdapter` per call. Caching was absent. 2. `_ensure_initialized()` was lazy — first tool call paid the index-build cost AND could race the index check on concurrent dispatch (e.g. preflight + bind landing in parallel after server boot). Together: every silent fallback in the production runtime was "hot" because the adapter was being rebuilt + rechecked on every call. Piece A (BicameralAI#283 commit 3c9730f) made the fallback loud at the response layer; Piece B closes the upstream cause. Three changes ------------- adapters/code_locator.py - Singleton-by-REPO_PATH cache via `_INSTANCE_CACHE: dict[str, RealCodeLocatorAdapter]`. Path resolved through `Path.resolve()` so symlink + relative-path callers cache-hit consistently. Multi-repo correctness preserved (any test that swaps REPO_PATH mid-process gets a fresh adapter for the new path). - New `reset_code_locator_cache()` test-only hook, mirroring `adapters.ledger.reset_ledger_singleton`. - New `async def RealCodeLocatorAdapter.initialize()` — wraps sync `_ensure_initialized()` in `loop.run_in_executor(None, ...)` so the cold-init path doesn't block the event loop. Idempotent on already-initialized adapters. server.py - `serve_stdio()` calls `await get_code_locator().initialize()` between the dashboard sidecar start and the consent-notice block. - **Fail-loud per BicameralAI#243 phase-2 signoff Q3** — explicit `except RuntimeError as exc:` re-raises after printing an actionable stderr message (`"Run: python -m code_locator index <repo>"`). The outer try/finally still runs the `SERVER_SHUTDOWN` audit emit, so operators get a clean event AND a clear actionable error. No more silent degradation. tests/test_preflight_graph_expansion.py — 4 new tests - test_get_code_locator_returns_same_instance_per_repo_path (singleton + reset behavior across two REPO_PATHs) - test_initialize_succeeds_when_index_present (idempotent on already-initialized adapter) - test_initialize_fails_loudly_when_index_empty (RuntimeError from `_ensure_initialized` propagates through the async wrapper — doesn't get swallowed) - test_serve_stdio_refuses_boot_on_empty_index (boot-path level: with everything else stubbed healthy, an empty index aborts `serve_stdio()` with the expected RuntimeError) Local smoke tests ----------------- - Singleton + reset_code_locator_cache: 4 assertions pass (cache hit on same path, distinct instance on new path, fresh after reset, second call after reset stays cached) - Async `initialize()`: re-raises RuntimeError on stubbed `_ensure_initialized` failure; idempotent no-op on already-initialized adapter - ruff check + ruff format --check + mypy all green on touched files What's NOT in this PR --------------------- Nothing — Piece A (commit 3c9730f) and Piece B (this commit) together close BicameralAI#243's full scope. PR will open with both pieces. Telemetry counter shipped in Piece A gives ongoing production visibility into how often fallback engages post-merge. Refs BicameralAI#243 (parent BicameralAI#173 / PR BicameralAI#174). Plan signoff via BicameralAI#243 (comment). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

jinhongkuan added enhancement New feature or request flow:feature Standard feature/fix PR targeting BicameralAI/dev (the default flow) P0 Critical: drop everything (prod down, data loss, security regression) preflight labels May 4, 2026

jinhongkuan temporarily deployed to production May 4, 2026 01:02 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to ci-test May 4, 2026 01:02 — with GitHub Actions Inactive

jinhongkuan had a problem deploying to recording-approval May 4, 2026 01:02 — with GitHub Actions Failure

jinhongkuan temporarily deployed to ci-test May 4, 2026 01:02 — with GitHub Actions Inactive

jinhongkuan mentioned this pull request May 4, 2026

feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal #173

Closed

6 tasks

jinhongkuan had a problem deploying to recording-approval May 4, 2026 01:17 — with GitHub Actions Failure

jinhongkuan temporarily deployed to production May 4, 2026 01:17 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to ci-test May 4, 2026 01:17 — with GitHub Actions Inactive

jinhongkuan had a problem deploying to recording-approval May 4, 2026 01:26 — with GitHub Actions Failure

jinhongkuan temporarily deployed to ci-test May 4, 2026 01:26 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to production May 4, 2026 01:26 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to ci-test May 4, 2026 01:26 — with GitHub Actions Inactive

jinhongkuan mentioned this pull request May 4, 2026

feat(skill): user-disambiguation question before Step 5.6 contradiction capture #175

Closed

4 tasks

jinhongkuan temporarily deployed to ci-test May 4, 2026 02:06 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to production May 4, 2026 02:06 — with GitHub Actions Inactive

jinhongkuan temporarily deployed to recording-approval May 4, 2026 02:06 — with GitHub Actions Inactive

jinhongkuan merged commit 8ca9d57 into dev May 4, 2026
8 checks passed

jinhongkuan added a commit that referenced this pull request May 4, 2026

Merge pull request #174 from BicameralAI/feat/preflight-graph-expansion

b3fb654

feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal

This was referenced May 4, 2026

feat(skill): add 'same intent' option to Step 5.6.1 disambiguation #176

Open

release: v0.13.7 (triage) #182

Merged

This was referenced May 7, 2026

Preflight: broaden dedup cache key to include file_paths + ledger revision (M7a/b/c) #87

Closed

feat(preflight): eliminate silent graph-expansion fallbacks (#173 follow-up) #243

Closed

This was referenced May 7, 2026

chore(v1): partial scale-down — revert preflight HITL bypass + decision_level wiring from dev (sibling to #242) #244

Closed

revert(v1): defer preflight HITL bypass + decision_level wiring (per #244) #246

Merged

This was referenced May 10, 2026

feat(preflight): eliminate silent graph-expansion fallbacks (#243) #294

Merged

M6 preflight handler retrieval: by-design split (handler structural, skill-layer covered by #306) #58

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal#174

feat(preflight): expand region-anchored lookup via 1-hop code-graph traversal#174
jinhongkuan merged 4 commits into
devfrom
feat/preflight-graph-expansion

jinhongkuan commented May 4, 2026

Uh oh!

coderabbitai Bot commented May 4, 2026 •

edited

Loading

Review skipped

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

jinhongkuan commented May 4, 2026

Summary

Linked issues

Plan / Audit / Seal

Test plan

Out of scope

Carry-over plan

Uh oh!

coderabbitai Bot commented May 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

coderabbitai Bot commented May 4, 2026 •

edited

Loading