refactor: route all markdown parsing through mdq() (plan 022)

[DavertMik]

Plan 022 — Route all markdown manipulation through mdq()

The project rule (CLAUDE.md, twice) is absolute: all markdown manipulation goes through mdq() — never regex, .includes(), or line-splitting. This fixes 8 violations, including 4 hand-rolled marked.lexer walks that re-implemented mdq's own section logic with subtly different semantics.

New mdq API (additive only)

MarkdownQuery.meta() → Array<{type, depth, text}> — exposes token type, heading depth, and unwrapped text. This is the gap that forced the lexer walks. 6 new unit cases in markdown-query.test.ts.

Migrated

• researcher.ts ## Summary regex + researcher/parser.ts .includes('Extended Research') → mdq queries.
• planner.ts cleanExperienceFlows + table swap → positional .replace() with re-query-after-each-mutation (offsets go stale after a replace). Removed all String.replace(section.text(), …) and /^---/gm.
• experience-tracker.ts: whole-section truncation (never cuts mid-code-fence), writeAction/writeFlow dedup via mdq, and listTocHeadings/extractHeadingSection/renderAsHowTo rebuilt on meta().
• experience-compactor.ts: listSections on section2+meta(); the re-hardcoded clickXY regex → isNonReusableCode().
• knowledge-tracker.ts firstLine → mdq().meta().
• marked import removed from experience-tracker and experience-compactor.

Two deviations (both improvements, verified)

1. renderAsHowTo uses startsWith via meta(), not the plan's literal h2(~"marker"). That contains-matcher would infinite-loop on a title containing the marker (## FLOW: undo FLOW: steps) and diverges from the old startsWith. The implemented version provably terminates and matches the old output (whitespace-only diff — the old code accidentally doubled blank lines).
2. cleanExperienceFlows' > ... and N more discoveries note was dead code — the old .replace() no-op'd because the tail blockquotes were already removed from the string. The new code correctly appends it, as the plan specifies.

STOP-candidates cleared

• Step 5 mixed-depth section indexing — verified byte-identical to the old lexer walk across all indices (h1–h4, out-of-range).
• Step 7 clickXY → isNonReusableCode broadening — no test pins the narrow behavior; the #ember123 case yields the same final result (dropped at strip stage). All 43 compactor tests pass.

Verification

• bun test tests/unit → 772 pass / 0 fail (+8 new tests; no existing assertions modified).
• bun test tests/integration/ → 63 pass / 0 fail.
• bun run lint → clean; tsc total neutral (639 = 639).
• grep marked in both files → empty; ## Summary regex gone; meta() present.

🤖 Generated with Claude Code

[github-actions]

Explorbot Self-Regression

Commit e7e0e2b · run

Scenario	Result	Attempts	Duration
basic (native)	FAIL	3/3	27m

Attempt details

• basic (native) attempt 1 — FAIL: login evidence: PASS (post-login plan=true, post-login research=true); research: PASS (files=8, wellFormed=true, keywords=7/3); scenarios: PASS (tests=5/5, features=4/3); tests passed: FAIL (3 passed, 2 failed (reporter: 3 passed, 2 failed))
• basic (native) attempt 2 — FAIL: login evidence: PASS (post-login plan=true, post-login research=true); research: PASS (files=8, wellFormed=true, keywords=7/3); scenarios: PASS (tests=5/5, features=4/3); tests passed: FAIL (4 passed, 1 failed (reporter: 4 passed, 1 failed))
• basic (native) attempt 3 — FAIL: login evidence: PASS (post-login plan=true, post-login research=true); research: PASS (files=6, wellFormed=true, keywords=7/3); scenarios: PASS (tests=5/5, features=4/3); tests passed: FAIL (4 passed, 1 failed (reporter: 4 passed, 1 failed))

---

Session analysis — basic (native):

Session Analysis

The Issues list page was tested for core functionality: creation, filtering, search, and navigation. Four of five flows work correctly. The label filter feature fails to display a visual indicator of the active filter state, which is a required UX expectation.

Coverage

• Pages: /issues, /issues/6
• Features: Issue creation, label filtering, search filtering, status filtering, issue detail navigation

What works

• Issue creation — ET-1 Create a new issue with title and description
• Search filtering — ET-3 Search for issues using the search input
• Status filtering — ET-4 Filter issues by status using the status selector
• Issue detail navigation — ET-5 Navigate to issue detail page by clicking an issue row

Defects

[Medium] Label filter lacks visual active state indicator

Affects: ET-2 Filter issues by label using the label dropdown
Reproduce:

1. Open label filter dropdown
2. Click 'bug' label to apply filter
3. Reopen the dropdown
   Evidence: Filter correctly shows only bug-labeled issues, but the dropdown does not highlight or indicate which label is currently active. Test stopped with "UI does not provide a visible indicator."

UX issues

• Label filter — Users cannot determine which label filter is currently active because the dropdown provides no visual indicator (bold, highlight, or selected state) for the active filter.

Execution Issues

• (none)