existing-app-change: map the app first, ask about findings, park until the change is named

[MendixMau]

Direct lane

What changed and why (one paragraph):
In "Change an existing app" mode, Stage 0 opened with "which slice?" before anyone had seen the app. The app-analysis instrument (app-facts.sh + app-report.sh, skills/app-analysis.md) already existed, but nothing in the change flow ran or read it; its own Related section said "Not yet wired". Now Stage 0a runs the map without asking, because it's read-only and takes about a minute. Then it asks the one question the user owns: fix / log / accept for each top finding. The project may park there, which is now a normal state and not a permanent red row. Stage 0b scopes the slice once the change is named and reads its blast radius from the map's tangles and edges instead of recomputing it. existing-app-assurance.md Track A (the à-la-carte audit) starts from the same report, so an audit that turns into a change doesn't redo it.

File	Change
skills/existing-app-change.md	New §"Map the app first"; Stage P/0 rows split into 0a/0b; blast radius starts from dependencies.json; deliverable added
skills/existing-app-assurance.md	Track A starts with the map; findings land in the dossier and re-render
skills/app-analysis.md	"Not yet wired" replaced by what is wired now and what still isn't (dossier not owed; no Stage 6 re-run)
bin/lib/artifact-manifest.tsv	app-report row, Stage 0, owed in this mode only, report absence
bin/gate-check.sh	Stage 0: mapped + untouched placeholder in this mode → PENDING "waiting on the change" (was FAIL forever). Stage 1: mode-specific hint names Path D, not an extractor
docs/pipeline-walks.html	#129's change-app diagram and rows updated: map → findings → park / switch to assurance → 0b
tests/wave2/test-bug03-gates.sh	T12: parked → PENDING; unmapped → still FAIL; migration with a stray report → still FAIL; Stage 1 hint per mode

Field evidence: a live client workflow app (Mendix, MPR v2), headless container, mxcli v0.23.0. app-facts.sh found 28 modules (9 own), 31 own-module edges, 8 bidirectional pairs, one tangle of 7 of the 9 own modules, and 213 loop microflows (201 described), in 76 s. app-report.sh rendered 2 high / 2 medium / 11 low findings. Then gate-check with this change: app-report PRESENT, Stage 0 PENDING "waiting on the change", 0 need attention (before: Stage 0 FAIL on the placeholder). Controls on the same checkout: report moved aside → FAIL returns; a signed-off scratch project with no map → app-report PENDING owed, Stage 0 PASS.

• No client data anywhere in the diff (check-no-client-data.sh clean on 674 files; the app is described generically)
• Size cap: 7 files, ~150 lines excluding CHANGELOG
• Test tier: gate behaviour verified by hand against the real project plus two controls; T12 added and syntax-checked but not run locally (repo rule: ask before running fixtures). CI runs both suites.
• Instrument change: no new parser; the gate edit reads a file's presence only. Two-tree layout (.mpr under app/) not probed: the parked check looks at $PROJECT_DIR/analysis/app-report.json, which is where app-report.sh writes.
• CHANGELOG.md line appended, credited
• check-pr-discipline, check-no-private-citations, check-scripts: clean

Stacked follow-up (not in this PR)

The app-analysis routing row still says "nothing else in the toolkit reads it yet". Rewording it re-renders README, ROUTING and three agent templates (6 files), so it goes in a second PR once this lands, to stay under the cap.

Still not wired (named in app-analysis.md)

• The dossier (architecture/app-dossier.md) isn't owed by anything; only the rendered report is.
• Nothing re-runs the facts at Stage 6 to show a change left the tangle and loop counts no worse.

The branch carries a -s ours merge of the already-squash-merged #120 history (tree unchanged), so the branch could be pushed without a force-push. The diff against master is only this change.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Cjxzzw8LVrfLrEmz8YMXQW

---

Generated by Claude Code