| name | verification-status |
| description | Render the per-task verification ledger (.hook-state/verification-ledger.json) — which gates ran and their outcomes — and record the manual checks CLAUDE.md mandates but a hook can't capture: the smoke-test result and silent-failure counts. Use before /ship or marking a task done, or when asked "did verification pass / what was verified". Do NOT use to run the gates (quality-gate.sh does that) or for install health (scripts/doctor.sh). |
Verification Status
Core Rule
A task isn't verified until the evidence is complete. The automatic gates (typecheck / lint) are logged by quality-gate.sh; the smoke test and silent-failure count are manual judgments a hook can't make. This skill shows the auto-gate evidence and records the manual half, so "would a staff engineer approve this?" is answerable from the ledger — not vibes.
When to Use
- Before
/ship or marking a task complete — assemble and show the verification evidence.
- When the user asks "did verification pass / what was verified / show verification status".
- Right after you run a manual smoke test — to record the result.
Do NOT use for:
- Running the gates —
quality-gate.sh does that automatically on every edit.
- Installation health / orphan hooks — that's
scripts/doctor.sh.
Process
-
Read the ledger .hook-state/verification-ledger.json. If it's absent, no qualifying edits ran this session — say so (nothing to verify yet).
python3 - "${CLAUDE_PROJECT_DIR:-$PWD}/.hook-state/verification-ledger.json" <<'PY'
import json, sys
try:
d = json.load(open(sys.argv[1]))
except FileNotFoundError:
print("No verification ledger — no qualifying edits this session."); raise SystemExit
for e in d.get("entries", []):
print(f" {e['status']:7} {e['tool']:24} {e.get('file','')} ({e.get('duration_s',0)}s)")
print(" smoke_test:", d.get("smoke_test"))
print(" silent_failures:", d.get("silent_failures"))
print(" coverage:", d.get("coverage"))
PY
-
Run the mandatory order if any step is missing (CLAUDE.md → Verification): typecheck → lint → tests → smoke test. The first three are auto-gated; you must run tests + the smoke test (open the page / call the endpoint / run the CLI).
-
Record the manual checks once you've done them — write into the ledger (atomic). Set the smoke result, the silent-failure tally (processed/failed/skipped), and optional coverage:
python3 - "${CLAUDE_PROJECT_DIR:-$PWD}/.hook-state/verification-ledger.json" \
"passed — opened /users, search returns results" \
"120/0/0" "" <<'PY'
import json, os, sys
f, smoke, silent, coverage = sys.argv[1:]
try:
d = json.load(open(f))
except FileNotFoundError:
d = {"schema_version": 1, "entries": [], "smoke_test": None,
"silent_failures": None, "coverage": None}
if smoke: d["smoke_test"] = smoke
if silent: d["silent_failures"] = silent
if coverage: d["coverage"] = coverage
tmp = f + ".tmp"
json.dump(d, open(tmp, "w"), indent=2); os.replace(tmp, f)
print("recorded")
PY
-
Verdict: state whether every mandated step is present and green. If a subset of items was silently skipped (failed/skipped > 0), surface it — never report "complete" over a silent drop.
Output Format
A short verification report:
- Auto-gates — table of
status · tool · file · duration from the ledger.
- Manual checks — smoke-test result; silent-failure tally
(processed/failed/skipped); coverage (if recorded).
- Verdict — one line: mandatory order complete & green? "Would a staff engineer approve this?"
Notes
quality-gate.sh writes the auto-gate entries; this skill reads them and records the manual half. The ledger is per-session (reset at session start) — it is evidence for the current task.stop-gate.sh emits a non-blocking reminder if auto-gates passed but no smoke result is recorded./ship cites this ledger as the verification evidence in its PR writeup.
