feat(fts5+mermaid): full FTS5 + Mermaid plan implementation (slices 1-7) + docs audit#69
Merged
SutuSebastian merged 3 commits intomainfrom May 4, 2026
Merged
Conversation
…ite path)
First slice of the FTS5+Mermaid plan implementation per
docs/plans/fts5-mermaid.md. Ships the FTS5 substrate end-to-end so
follow-up slices (demo recipe, Mermaid formatter, MCP/HTTP plumbing)
have something to layer on.
Schema (per Q1):
- SCHEMA_VERSION 6 → 7 (forces dropAll on first upgrade).
- New `source_fts` virtual table — columns (file_path UNINDEXED,
content), tokeniser 'porter unicode61'. Always created; near-zero
space when empty.
- Helpers: upsertSourceFts (DELETE+INSERT — FTS5 doesn't accept
INSERT OR REPLACE on virtual tables), deleteSourceFts (manual
mirror of FK CASCADE since virtual tables can't be FK targets),
clearSourceFts.
Config (per Q6):
- New `fts5: boolean` field in Zod schema (default false, optional).
- `ResolvedCodemapConfig.fts5` resolved from config + CLI; CLI wins
per `--root` / `--state-dir` precedent. Logs stderr override line
when CLI overrides config.
- New `getFts5Enabled()` runtime accessor.
CLI:
- `--with-fts` flag in parseBootstrapArgs; threaded through
bootstrapCodemap → resolveCodemapConfig.fts5Cli; rest also pushed
into validateIndexModeArgs allowlist.
Worker plumbing (per Q2):
- WorkerInput.fts5Enabled propagated from worker-pool via
getFts5Enabled().
- ParsedFile.content optional; worker tees source into it only when
fts5Enabled (zero serialization cost on default-OFF path).
Indexer (per Q2):
- insertParsedResults writes source_fts in same transaction as files
row insert.
- Single-threaded path (parse-on-main, used in incremental
per-relPath loop) also calls upsertSourceFts gated on
getFts5Enabled().
- deleteFilesFromIndex mirrors DELETE to source_fts.
- indexFiles fullRebuild path re-seeds meta (fts5_enabled,
schema_version) after dropAll wipes meta.
Toggle-change auto-detect (per Q3):
- run-index reads meta.fts5_enabled; mismatch with current resolved
config upgrades incremental → full and logs stderr line. First-run
(no value) seeds silently. Already-full mode skips the upgrade
message but still syncs.
Verification:
- bun run typecheck passes.
- bun test: 746 pass, 0 fail.
- End-to-end smoke test (/tmp/fts5-smoke-test):
- Index without --with-fts → source_fts empty ✓
- Re-index with --with-fts → toggle-change auto-detect upgraded
incremental to full → MATCH 'TODO' returns the seeded file ✓
Slice 1 deliberately does NOT include:
- Telemetry on first FTS5 populate (Q7) — defer.
- Mermaid formatter (slices 3-5).
- Demo recipe text-in-deprecated-functions (slice 2).
- Docs/agents lockstep (slice 6) — defer until the demo recipe lands
so the rule update names a real recipe.
- Patch changeset (slice 7).
Files changed: 9 (db.ts, config.ts, runtime.ts, bootstrap.ts,
bootstrap-codemap.ts, main.ts, cmd-index.ts, parsed-types.ts,
parse-worker-core.ts, worker-pool.ts, index-engine.ts, run-index.ts).
🦋 Changeset detectedLatest commit: 6cfa3f5 The changes in this PR will be included in the next version bump. This PR includes changesets to release 1 package
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
|
Warning Rate limit exceeded
To keep reviews running without waiting, you can enable usage-based add-on for your organization. This allows additional reviews beyond the hourly cap. Account admins can enable it under billing. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (29)
✨ Finishing Touches🧪 Generate unit tests (beta)
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. Review rate limit: 0/1 reviews remaining, refill in 9 minutes and 52 seconds.Comment |
…try + lockstep Completes the FTS5+Mermaid plan (docs/plans/fts5-mermaid.md). Builds on slice 1's substrate. Slice 2 — demo recipe text-in-deprecated-functions: Bundled recipe: @deprecated functions in files containing TODO/FIXME/ HACK markers AND coverage <50%. Demonstrates FTS5 ⨯ symbols ⨯ coverage JOIN composability that ripgrep can't match. Returns empty when FTS5 is off (source_fts empty). Action template: review-cleanup-priority. Slices 3+4 — Mermaid formatter with bounded-input contract: - formatMermaid(rows, opts) in output-formatters.ts. Renders {from, to, label?, kind?} as flowchart LR. Reuses existing formatter plumbing pattern from SARIF / annotations. - MERMAID_MAX_EDGES = 50 hard-coded const (Q4). Auto-truncation explicitly out of scope (would be a verdict masquerading as output mode, violating moat A). - Reject error names recipe + count + scoping knobs (LIMIT / --via / WHERE) so agent knows how to scope. Slice 5 — MCP + HTTP plumbing: - formatEnum gains "mermaid"; QueryArgs / QueryRecipeArgs format unions extended. - tool-handlers.runFormattedQuery branches on "mermaid" → wraps formatMermaid in try/catch (bounded-input rejection becomes a structured ToolResult error). - MCP wrapToolResult needs no change — generic non-JSON passthrough handles "mermaid" same as "sarif" / "annotations". - HTTP writeToolResult needs no change — fall-through "text/plain; charset=utf-8" applies to mermaid (not sarif+json). - Tool descriptions in mcp-server.ts updated to mention "mermaid" format. Slice 6 — agent rule + skill lockstep (Rule 10): Both templates/agents/ AND .agents/ codemap rule + skill updated: - --format mermaid example row in CLI table - --with-fts row in CLI table - text-in-deprecated-functions in trigger pattern + recipe id list Slice 7 — Q7 telemetry + patch changeset: - Stderr line on first FTS5 populate: [fts5] source_fts populated: <N> files / <X> KB (uncompressed content). Cheap (single SUM(length(content)) on source_fts at end of full reindex); only fires when fts5 just became populated (fts5WasEmpty pre-check). - Patch changeset (.changeset/fts5-mermaid.md) per pre-v1 lesson: additive feature, default-OFF, behaviour-preserving for existing users. Verification end-to-end: - bun test: 754 pass, 0 fail (8 new mermaid formatter tests) - bun run check passes (format, lint, typecheck, 23/23 golden queries) - Smoke test with @deprecated function + TODO comment: * codemap --with-fts --full → "[fts5] source_fts populated: 2 files / 138 B (uncompressed content)" ✓ * query --recipe text-in-deprecated-functions → returns the deprecated function ✓ * query --format mermaid 'SELECT from_path AS "from", to_path AS "to" FROM dependencies LIMIT 50' → renders flowchart LR ✓ * query --format mermaid 'SELECT from_path AS "from", to_path AS "to" FROM dependencies' (unbounded) → rejects with scope- suggestion error ✓ Plan slices all complete. PR #69 description gets refreshed in a follow-up commit (this commit covers the impl — the doc-audit pass on docs/ comes next per user request).
Per docs-governance lifecycle (docs/README.md Rule 3 + Lifecycle table:
"Plan: Deleted when work ships"):
LIFTED (decisions promoted from docs/plans/fts5-mermaid.md to canonical
homes):
- docs/architecture.md
- Schema version 6 → 7
- meta table description: added fts5_enabled key
- New "source_fts" table section under § Schema, after meta —
documents tokenizer, file_path UNINDEXED, opt-in toggle precedence,
auto-detect, telemetry, and the bundled demo recipe
- application/ engines list: output-formatters.ts now mentions
Mermaid alongside SARIF + GH annotations
- docs/glossary.md
- New "source_fts (FTS5 virtual table) / --with-fts / opt-in
full-text" entry — covers schema shape, toggle precedence,
auto-detect, telemetry, default-OFF rationale, the JOIN
composability the FTS5 / ripgrep distinction tracks
- New "--format mermaid / formatMermaid / MERMAID_MAX_EDGES" entry
— covers the {from, to, label?, kind?} input contract, bounded-
input rejection (50 edges), why auto-truncation is out of scope
- docs/roadmap.md
- Backlog: "Optional FTS5 for opt-in full-text" line removed (work
has shipped per Rule 2 — backlog items move OUT when shipped)
- Non-goals: "Full-text search across all file bodies — use ripgrep
/ IDE / opt-in FTS5 (see backlog)" REWORDED to "Full-text search
default-on" non-goal — FTS5 ships per --with-fts, but default-on
is still out of scope until v2 size-tax measurements
DELETED:
- docs/plans/fts5-mermaid.md (per Rule 3 + Lifecycle: plans deleted
when work ships; decisions live in canonical homes above)
- docs/README.md File Ownership row updated to drop fts5-mermaid.md
from in-flight plans list (still has c9-plugin-layer.md)
Audit method: grep [Ff]allow / FTS5 / source_fts / --with-fts /
--format mermaid across docs/ — categorised hits as documentation
(lift to architecture.md / glossary.md), historical (research notes
keep their content; closed records like fallow.md stay closed), or
plan-residue (delete fts5-mermaid.md per lifecycle).
Verification:
- Schema version assertion in architecture.md matches SCHEMA_VERSION
in db.ts (7).
- meta.fts5_enabled key documented matches META_FTS5_ENABLED_KEY in
db.ts.
- source_fts column shape (file_path UNINDEXED, content) +
tokenizer ('porter unicode61') match the actual CREATE VIRTUAL
TABLE in db.ts createTables.
- --with-fts CLI flag documented matches parseBootstrapArgs +
validateIndexModeArgs allowlists.
- --format mermaid documented matches OUTPUT_FORMATS + formatEnum in
cmd-query.ts + tool-handlers.ts.
- 50-edge ceiling matches MERMAID_MAX_EDGES in output-formatters.ts.
- bun run check passes (format, lint, typecheck, 23/23 golden
queries).
SutuSebastian
changed the title
4 tasks
SutuSebastian
added a commit
that referenced
this pull request
May 4, 2026
… on parser.ts Fact-checking against codebase post-PR-#69-and-#70 surfaced four stale spots; concise-comments rule re-applied to recently-authored parser.ts comments. DOCS LIFTED (post-FTS5 / Mermaid / complexity merge): - README.md (root) line 113 — --format enum was missing `mermaid`. Updated to <text|json|sarif|annotations|mermaid> + added the bounded-input contract one-liner + 50-edge ceiling note. Added --with-fts example block alongside (was missing entirely; README is the canonical CLI surface per docs/README.md Single source of truth table). - docs/architecture.md output-formatters paragraph — described only formatSarif + formatAnnotations; missing formatMermaid + bounded- input contract. Added formatMermaid description + MERMAID_MAX_EDGES reference + the no-auto-truncation reasoning (would be a verdict masquerading as output mode). Updated the --format CLI enum to include mermaid; same for the MCP tools format union. - .agents/skills/codemap/SKILL.md + templates/agents/skills/codemap/ SKILL.md — recipe-id list missed three coverage recipes (untested-and-dead, files-by-coverage, worst-covered-exports) shipped earlier in PR #65/#56 era. Lockstep update per Rule 10. Skill now lists 20 of 20 bundled recipe ids. CONCISE-COMMENTS SWEEP on parser.ts (recently authored): - Trimmed the 14-line complexityStack JSDoc block to 6 lines. Kept: the -1 sentinel rationale (non-obvious), the WeakMap rationale (the bug fix from PR #70 review). Cut: re-stating push/pop semantics obvious from method names + step-by-step "this then that" prose. - Removed the "Defer complexity push to..." comment in the VariableDeclaration handler. The 4-line block restated the design decision documented one screen up in the complexityStack jsdoc; cross-ref makes it redundant. Per concise-comments § "Cut" rule: "Cross-references that save grep time" — keep when they actually do; cut when they restate. Verification: - bun run check: format + lint + typecheck + 23/23 golden ✓ - Recipe count: SQL files = 20, skill mentions = 20 (1:1 match) ✓ - SCHEMA_VERSION = 8 in db.ts; docs/architecture.md says 8 ✓ - complexity column documented in architecture.md + glossary.md ✓ - --with-fts in README.md + architecture.md + glossary.md + roadmap.md (consumer-facing surfaces all aligned) ✓ - --format mermaid in README.md + architecture.md + glossary.md + agent rule/skill ✓
SutuSebastian
added a commit
that referenced
this pull request
May 4, 2026
…sted recipe (research note § 1.4) (#70) * feat(complexity): cyclomatic complexity column + high-complexity-untested recipe Research note § 1.4 ship-pick (c) per § 5 cadence. Schema bump SCHEMA_VERSION 7 → 8. Schema: - symbols.complexity REAL column. NULL for non-function kinds and class methods (v1 limitation documented in recipe .md). Parser: - complexityStack maintained alongside scopeStack. Function entry pushes {symbolIndex, count: 1}; branching-node visitors increment top.count; function exit pops + writes count into the already- pushed symbol row's complexity field. - McCabe decision points counted: if, while, do-while, for, for-in, for-of, case X (not default:), &&/||/??, ?:, catch. Bundled recipe high-complexity-untested: - Joins symbols (complexity >= 10) with coverage (< 50%). - Combines structural + runtime evidence axes — surfaces refactor- priority candidates that untested-and-dead and worst-covered-exports miss (they catch dead-or-uncalled, this catches called-but-undertested- AND-branchy). Empirical sanity check on codemap's own index after reindex: - extractFileData (parser.ts main visitor) → complexity 108 ✓ - stringifyTypeNode → 42 ✓ - All non-function kinds have NULL complexity ✓ - high-complexity-untested recipe returns 7 functions all from src/parser.ts (which has 0% coverage; complexity ≥ 10) ✓ Lockstep updates per Rule 10 (templates/agents + .agents): - Trigger pattern row "What's high-complexity AND undertested?" - Quick reference row for SELECT name, complexity FROM symbols - Recipe-id list extended in SKILL.md Plus architecture.md (schema version 8, complexity column docs), glossary.md (cyclomatic complexity entry), patch changeset. Files changed: - src/db.ts (SCHEMA_VERSION + symbols.complexity column + insertSymbols bind + SymbolRow optional complexity field) - src/parser.ts (complexityStack + branching node visitors + push/pop in FunctionDeclaration / VariableDeclaration arrow-fn paths) - templates/recipes/high-complexity-untested.{sql,md} - docs/architecture.md (schema version + symbols column doc) - docs/glossary.md (new entry) - templates/agents/rules/codemap.md + .agents/rules/codemap.md (trigger + quick-ref rows) - templates/agents/skills/codemap/SKILL.md + .agents/skills/codemap/ SKILL.md (recipe-id list) - .changeset/cyclomatic-complexity.md (patch) Verification: - bun test: 754 pass - bun run check passes (format, lint, typecheck, 23/23 golden queries) - Live re-index against codemap source produces sensible complexity values (parser visitor itself is the highest at 108, which tracks) * docs(skill): add complexity column to symbols schema in skill files CodeRabbit catch on PR #70: the high-complexity-untested recipe row was added to .agents/skills/codemap/SKILL.md but the symbols table schema section (under "### `symbols` — Functions, types, ...") still listed columns through `visibility` only, missing the new `complexity REAL` column. Verified by reading the file — claim was correct. Both lockstep mirrors (.agents/ + templates/agents/) updated with the same row: | complexity | REAL | Cyclomatic complexity (`1 + decision points`) for function-shaped symbols. NULL for non-functions and class methods (v1). Powers --recipe high-complexity-untested. Decision points: if, while, do…while, for/for-in/for-of, case X: (not default:), &&/||/??/?:, catch | Per docs/README.md Rule 10 — agent rule + skill schema docs must stay in lockstep with code-side schema changes. The trigger-pattern row + recipe-id list were already updated; the schema-table row was the gap. * fix(complexity): per-function visitors fix multi-declarator misattribution + cleanups CodeRabbit raised three valid findings on PR #70. All fact-checked against the code; all correct. A) docs/architecture.md symbols schema table was malformed: - Markdown table separator row had extra `| --- | ---` segments because oxfmt mis-counted columns when the description contained `|` chars inside `&&`/`||`/`??` backtick spans. - The complexity row's description was split across THREE cells with broken backtick fences. - Fix: restored single-row layout (3 cells: Column | Type | Description) and rephrased the decision-point list to avoid `|` inside backticks ("short-circuit `&&` / `||` / `??`" instead of "`&&`/`||`/`??`"). B) src/parser.ts complexity misattribution on multi-declarator VariableDeclaration (e.g. `const a = () => {…}, b = () => {…};`): Pre-fix: VariableDeclaration enter pushed all declarators' complexity entries up front. Then visitor traversed `a`'s body — branches incremented top (= b's entry). Then `b`'s body. Exit pops in reverse → symbols[1].complexity = 3 (wrong), symbols[0].complexity = 1 (wrong). Real bug. Fix: push/pop complexity on the FUNCTION-shaped node visitors (ArrowFunctionExpression / FunctionExpression) — not on VariableDeclaration. The VariableDeclaration handler still creates the symbol row but only RECORDS the symbol → init-node mapping in a WeakMap. The ArrowFunctionExpression / FunctionExpression enter handler reads the WeakMap to know which symbol to write back to; anonymous arrow fns (callbacks, IIFEs) get -1 and just track count without persistence. Verified against fixture: const a = () => { if (1===1) {…} }, b = () => { if (2===2) {…} }, c = () => 5; → a=2, b=2, c=1 (correct; pre-fix was a=1, b=3, c=1) C) popComplexityInto guard was a no-op (callers passed top.symbolIndex, so the equality check was always true). Simplified to parameterless popComplexityTop() that always pops + writes back if symbolIndex >= 0. Folds naturally into the B refactor — every push/pop pair now lives in a function-shaped visitor. Also re-ran codemap query against codemap source post-fix: extractFileData=108, stringifyTypeNode=42, extractClassMembers=18, extractLiteralValue=15, extractObjectMembers=14 Same scores as pre-fix on these (no FunctionExpression / arrow nesting in those particular functions, so the bug didn't surface) — confirms the refactor is a strict improvement, not a regression. * docs: audit + lift remaining stale references; concise-comments sweep on parser.ts Fact-checking against codebase post-PR-#69-and-#70 surfaced four stale spots; concise-comments rule re-applied to recently-authored parser.ts comments. DOCS LIFTED (post-FTS5 / Mermaid / complexity merge): - README.md (root) line 113 — --format enum was missing `mermaid`. Updated to <text|json|sarif|annotations|mermaid> + added the bounded-input contract one-liner + 50-edge ceiling note. Added --with-fts example block alongside (was missing entirely; README is the canonical CLI surface per docs/README.md Single source of truth table). - docs/architecture.md output-formatters paragraph — described only formatSarif + formatAnnotations; missing formatMermaid + bounded- input contract. Added formatMermaid description + MERMAID_MAX_EDGES reference + the no-auto-truncation reasoning (would be a verdict masquerading as output mode). Updated the --format CLI enum to include mermaid; same for the MCP tools format union. - .agents/skills/codemap/SKILL.md + templates/agents/skills/codemap/ SKILL.md — recipe-id list missed three coverage recipes (untested-and-dead, files-by-coverage, worst-covered-exports) shipped earlier in PR #65/#56 era. Lockstep update per Rule 10. Skill now lists 20 of 20 bundled recipe ids. CONCISE-COMMENTS SWEEP on parser.ts (recently authored): - Trimmed the 14-line complexityStack JSDoc block to 6 lines. Kept: the -1 sentinel rationale (non-obvious), the WeakMap rationale (the bug fix from PR #70 review). Cut: re-stating push/pop semantics obvious from method names + step-by-step "this then that" prose. - Removed the "Defer complexity push to..." comment in the VariableDeclaration handler. The 4-line block restated the design decision documented one screen up in the complexityStack jsdoc; cross-ref makes it redundant. Per concise-comments § "Cut" rule: "Cross-references that save grep time" — keep when they actually do; cut when they restate. Verification: - bun run check: format + lint + typecheck + 23/23 golden ✓ - Recipe count: SQL files = 20, skill mentions = 20 (1:1 match) ✓ - SCHEMA_VERSION = 8 in db.ts; docs/architecture.md says 8 ✓ - complexity column documented in architecture.md + glossary.md ✓ - --with-fts in README.md + architecture.md + glossary.md + roadmap.md (consumer-facing surfaces all aligned) ✓ - --format mermaid in README.md + architecture.md + glossary.md + agent rule/skill ✓
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
Complete implementation of the FTS5+Mermaid plan (
docs/plans/fts5-mermaid.md, now deleted per docs-governance lifecycle). All seven slices in one PR; decisions lifted to canonical homes (architecture.md,glossary.md,roadmap.md).What ships
FTS5 (opt-in, default OFF)
source_ftsvirtual table —(file_path UNINDEXED, content)columns,tokenize='porter unicode61'. Always created (near-zero space when empty).codemap.config.tsfts5: trueOR--with-ftsCLI flag at index time. CLI wins; logs stderr line on override.parsed.contentteed from worker only when toggle on (zero serialization cost on default-OFF). Atomic withfilesrow insert.meta.fts5_enabledmismatch upgrades incremental → full sosource_ftsis consistently populated.[fts5] source_fts populated: <N> files / <X> KB.text-in-deprecated-functions—@deprecatedfunctions in files withTODO/FIXME/HACKmarkers AND coverage<50%. Demonstrates the FTS5 ⨯symbols⨯coverageJOIN.Mermaid output formatter
--format mermaidrendering{from, to, label?, kind?}rows asflowchart LR.MERMAID_MAX_EDGESconst). Unbounded inputs reject with scope-suggestion error naming recipe + count +LIMIT/--via/WHEREknobs.query/query_recipe+ HTTPPOST /tool/query— same plumbing as SARIF/annotations.Schema bump
SCHEMA_VERSION6 → 7. First reindex after upgrade triggers a full rebuild via the existing version-mismatch path. Existing user data (query_baselines,coverage) preserved (intentionally absent fromdropAll()).Docs audit (lift + delete)
Per docs-governance Rule 3 + Lifecycle ("Plan: Deleted when work ships"), decisions lifted to canonical homes:
source_ftsschemadocs/architecture.md§ Schema (new section)--with-ftsCLI flagdocs/architecture.md§ Schema cross-ref + glossary entrymeta.fts5_enabledsemanticsdocs/architecture.md§ meta table description--format mermaidshapedocs/architecture.md§ application/ engines list + glossary entrydocs/glossary.md--format mermaidentrydocs/glossary.mdsource_ftsentrydocs/roadmap.mdBacklog: removed (Rule 2 — shipped items move out)docs/roadmap.mdNon-goals: reworded to "default-on" non-goalPlus
docs/plans/fts5-mermaid.mddeleted;docs/README.mdFile Ownership updated.Verification
End-to-end smoke test:
Test plan
bun test— 754 / 754 pass (8 new Mermaid formatter tests)bun run check— format + lint + typecheck + 23/23 golden queriesarchitecture.mdmatchesSCHEMA_VERSIONconst indb.tsmeta.fts5_enableddocumented matchesMETA_FTS5_ENABLED_KEYconstMERMAID_MAX_EDGESconstPre-v1 patch changeset
Per
.agents/lessons.md"changesets bump policy" — additive feature, default-OFF for FTS5, behaviour-preserving for existing users (--with-ftsopt-in; Mermaid is a new output mode, not a breaking change to existing formats).Doc-governance compliance
templates/agents/AND.agents/codemap rule + skill mention--with-fts,--format mermaid,text-in-deprecated-functionsrecipe, bounded-input contract.concise-comments— comments swept post-write; kept only non-obvious-constraint commentary (FTS5 no-INSERT-OR-REPLACE, FK CASCADE doesn't reach virtual tables, dropAll wipes meta) + spec cross-refs (Q1/Q2/Q3/Q6).Out of scope
Other research note § 1 candidates remain parked: § 1.3 cyclomatic complexity (AST walker), § 1.5 boundary violations (Zod config + new table), § 1.6 unused type members (substrate gap), § 1.9 recipe-recency (new table + reconciler), § 1.10 rename-preview (parametrised-recipes infra). The (b) C.9 plugin layer plan continues iterating in parallel per the original ship sequence.