Cover the pin-shortcut run path and the job self-improvement footer

[selfcontained]

Recurring test-health run. Local suite was green on the first pass (28th consecutive), so the whole run went to the queued next_focus coverage gap.

Suite status

Lane	Result
pnpm run check	pass
pnpm run format	pass on a warm tree (#942's .prettierignore fix still holding)
pnpm run test	server 162 files / 2735 tests, web 912, ext 60, scripts 4
pnpm run test:e2e	180 passed / 12 skipped
pnpm run test:e2e:live	11 passed in 23.3s, zero leaked e2e-* tmux sessions (19th consecutive)
pnpm run finalize:web	pass

CI scan (gh run list --limit 50): 49 success / 1 failure. The failure is the Release workflow's own origin/main advanced after smoke tests merge-race guard on v0.33.12 from 2026-08-13 — already triaged in a prior run, not a test failure and not a flake. No new flakes; the brain flake list stays empty (6th consecutive run).

Coverage added

Three zero-coverage surfaces, all reached through the queued focus area:

• useRunPinShortcut (apps/web/src/hooks/use-pin-shortcuts.ts) had no unit coverage at all. It is a write into a live agent session — the server resolves the pin's prompt and types it into the terminal. The E2E suite proves the button reaches the endpoint once; the URL shape, the POST, and the label/error toast branches are only checkable at this level.
• resolvePinShortcutIcon (apps/web/src/lib/pin-shortcut-icons.ts) was covered for its name allowlist by the existing server-side lockstep test, but not for resolution. The lookup is an own-property check specifically so inherited keys fall through to the fallback — without it, an agent passing "constructor" resolves to something that is not a component and throws during render.
• buildSelfImprovementGuidance's job branch (apps/server/src/shared/self-improvement-prompt.ts) had no assertions anywhere. jobs/service.ts builds its prompt in a module-private function, so nothing else in the repo checks that a self-improving job is told to call update_job with its own name and directory. The template branch already reaches three suites through its caller and is deliberately not duplicated here.

No product code changed. No product bugs found, no Linear tickets filed.

Mutation battery

21 mutants, 19 killed. Two survivors, neither a test gap:

• resolvePinShortcutIcon's name && truthiness guard is unobservable — Object.hasOwn already rejects undefined (coerced to the absent key "undefined") and "". Recorded as product cleanup, not missing coverage.
• Splitting the update instruction across two lines changes guidance formatting rather than any contract. Protecting it meant pinning a 155-char prose string that the self-improvement flow is itself meant to be able to reword, so it was deliberately left unprotected.

Review round

One general review agent inspected this run's diff and returned five actionable items, all applied:

1. Deleted a tautological loop asserting resolvePinShortcutIcon(k) === PIN_SHORTCUT_ICONS[k] for every own key — that restates the implementation, and the typeof companion was already guaranteed by the satisfies Record<string, LucideIcon> annotation.
2. Dropped the prose-pinned one-line assertion (see the second surviving mutant above).
3. Dropped the template-branch test — the review found it already covered by template-launch-prompt.test.ts, templates/service.test.ts, and mcp-handlers.test.ts. Re-running the battery confirmed nothing was lost: the templateId mutant is still killed by the existing suites.
4. Dropped three assertions that certified react-query rather than the hook (toHaveBeenCalledTimes(1) after a single mutate, and the two "the other toast didn't fire" cross-checks).
5. Fixed a misattributed comment — apps/web pins vitest 2.1.9, and restoreAllMocks has never reset module-factory vi.fn()s in either major, so the justification now names the actual reason rather than a version.

🤖 Generated with Claude Code