feat(scaffold): migration 0007 ignores .taskless/.tmp-* scratch files

[thecodedrift]

Stack (root → tip):

• docs(openspec): propose qualitative survey feedback via PostHog surveys #341
  • ➡️ feat(scaffold): migration 0007 ignores .taskless/.tmp-* scratch files #342 (you are here)
    • feat(feedback): add feedback send/dismiss, the survey constants, and the cadence store #343
      • docs(agent): add the feedback and feedback-invite recipes #344
        • feat(agent): append the feedback survey invite to surveyed recipes #345

Slice 1 of 4 of qualitative-survey-feedback (stacked on #341, the proposal). Bottom of the implementation stack; the changeset lives here.

What

• 0007-ignore-scratch-files.ts: addToGitignore(cwd, ["/.tmp-*"]), registered as "7". 0001 is not edited; a fresh scaffold runs 1 through 7 and ends with the same .gitignore an upgraded one has.
• Anchored with a leading / for the reason 0001 anchors /sgconfig.yml: an unanchored pattern would match inside a rule directory.
• This repository's own scaffold migrated (.taskless/.gitignore gains the line, taskless.json records 7).
• Changeset: patch. The scaffold's own version field carries the compatibility signal.

Tests

test/migrate-ignore-scratch-files.test.ts: version-6 scaffold gains the line and records 7 with existing entries intact; fresh scaffold ends with all three entries; migration is idempotent; and git check-ignore confirms a .taskless/.tmp-feedback.json is ignored (asks git rather than re-deriving its rules). The existing forward-migration matrix picks up v6 → v7 automatically since it derives from LATEST_SCHEMA_VERSION.

Refs #341

[thecodedrift]

@claude /review

[claude]

Claude finished @thecodedrift's task in 3m 19s —— View job

---

Review mode: incremental — read 0 prior review thread(s) before reviewing.

.prior-review.json contained no review threads, review summaries, or top-level comments beyond the @claude /review trigger itself, so there is nothing previously raised to classify as addressed or still-open. This is effectively the first substantive review of this PR.

• Read .prior-review.json
• Fetch PR diff and metadata (gh pr diff 342, gh pr view 342)
• Review changed files in full context (traced Migration call sites, addToGitignore, runMigrations, sibling migrations 0001/0004/0006, the test matrix, and the canonical skill source)
• Post findings

Assessment

This is a clean, well-scoped slice. Walked through the whole chain and didn't find a correctness bug:

• 0007-ignore-scratch-files.ts: Migration is invoked with the .taskless directory (see runMigrations in migrate.ts:494, await migrate(tasklessDirectory)), and dirname(directory) correctly recovers the project root — the same result 0001-init.ts gets via join(directory, ".."). addToGitignore then re-derives .taskless internally, so the round trip is correct and matches the established pattern.
• Anchoring: /.tmp-* anchors to .taskless/ itself, consistent with 0001's /sgconfig.yml reasoning — an unanchored pattern would also match a rule directory's own file of that name.
• Registration: migrate.ts registers "7": ignoreScratchFiles and leaves 0001 untouched, matching the PR description. LATEST_SCHEMA_VERSION is derived from the migration map, so the existing migrate-install.test.ts matrix picks up 6→7 automatically, as claimed.
• Idempotency / fresh-install / upgrade paths: all three are exercised in migrate-ignore-scratch-files.test.ts, and the "asks git rather than re-deriving its rules" git check-ignore test is a good way to verify the actual guarantee that matters (a real scratch file becomes invisible to git status), consistent with this repo's existing pattern of shelling out to real git in tests (check-gitignore.test.ts, check.test.ts, etc.).
• Self-hosting: this repo's own .taskless/.gitignore and taskless.json (version: 7) were updated in the same commit, per the "a migration owns everything under .taskless/" rule documented in migrate.ts.
• Changeset: patch, placed on this bottom-of-stack PR — correct under the 0.y.z "added surface is still patch" rule in the repo's CLAUDE.md, and correctly placed per the stacked-PR changeset convention.
• Unrelated-looking second commit (.agents/skills/taskless/SKILL.md): this drops a hardcoded npx @taskless/cli prefix from the frontmatter description's agent route reference. Verified this brings the stub back in sync with the canonical skills/taskless/SKILL.md and the other installed stub copy (.claude-pr/.claude/skills/taskless/SKILL.md), both of which already read plain `agent route` with no prefix — so this is a real drift fix, not noise, and it's a good example of the exact frontmatter-vs-body distinction the commit message draws (the body's npx @taskless/cli init recovery line is a different, intentionally-frozen mechanism handled by stubRecoveryInvocationStale, and is correctly left alone here).

Nothing NEW to flag beyond the above observations, which are confirmations rather than issues. No inline comments filed.

[thecodedrift]

Re: @claude[bot] — "Claude finished @thecodedrift's task in 3m 19s … Nothing NEW to flag beyond the above observations"
#342 (comment)

Read; no findings to act on. The stub refresh commit is intentional and is described in its own message.

— AI Coding Agent