test(gates): pin hook wiring and enforce the bats TAP plan

The repo's gates were not pinned to their wiring. Deleting the
check-skill-version-bump block from .pre-commit-config.yaml left the
whole suite green; deleting eight blocks at once, run-tests among them,
also left it green. Only 4 of 20 hook ids had their wiring pinned
anywhere, so a merge conflict resolved badly could stop the suite
running at pre-push forever while every test still reported green.

test-adr0020-contract now derives the repo-authored hooks from the
repo: local entries and pins each one's id, entry and stages against an
explicit expected set, both directions, with the same non-vacuity
guards the file already applies to its own fixtures. Upstream hooks and
their rev: values are untouched, so a rev bump does not churn the test.
Mutation-checked: a removed block, a repointed entry and a hook moved
off pre-push each go red; a rev bump, a comment edit and reordering
stay green. 29 -> 44 assertions.

run-bats computed each file's TAP plan and then discarded it, so a
process printing "1..10", three ok lines and exit 0 was counted as
"3 tests, 0 failures" with seven tests silently gone. That is exactly
the wrapper-swallows-the-status case the runner's own comment puts in
its threat model, and the plan was the only surviving signal. The plan
is now enforced in both directions when a file emits exactly one.

Also: test-no-pipefail-early-exit-grep's live-tree floor goes from 20 to
50 against an actual 57, matching test-vale-wrap's per-glob discipline,
and test-vale-wrap's header names the real path to vale-wrap.sh.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NwD8Egs5r4ndqeFLmhusX2
This commit is contained in:
2026-09-20 12:34:07 +00:00
parent 8cfd54f925
commit 384756b343
5 changed files with 271 additions and 10 deletions

View File

@@ -180,6 +180,15 @@ for f in ${TEST_FILES[@]+"${TEST_FILES[@]}"}; do
# file bats really did run, so it has to be distinguishable from a file that
# produced nothing whatsoever.
file_plan="$(grep -c '^1\.\.[0-9]' "$SCRATCH_ROOT/$i.log" || true)"
# The planned count itself, extracted only when the file emitted exactly one
# plan line. Zero plans is the broken-harness case the aggregate guard below
# names, and two or more means the log is not one file's TAP stream at all --
# in neither case does "the planned count" mean anything, so the per-file
# comparison is skipped and the previous behaviour stands.
file_planned=""
if [[ "$file_plan" -eq 1 ]]; then
file_planned="$(sed -n 's/^1\.\.\([0-9][0-9]*\).*$/\1/p' "$SCRATCH_ROOT/$i.log")"
fi
# String-compared below, not `-ne`. `-ne` is arithmetic and bash evaluates an
# empty string as 0 there -- `[[ "" -ne 0 ]]` is false -- so an *empty* status
# file read as a clean exit. The `|| echo 1` fallback only covers a *missing*
@@ -189,14 +198,28 @@ for f in ${TEST_FILES[@]+"${TEST_FILES[@]}"}; do
TOTAL_OK=$((TOTAL_OK + file_ok))
TOTAL_NOT_OK=$((TOTAL_NOT_OK + file_not_ok))
TOTAL_PLANS=$((TOTAL_PLANS + file_plan))
# Two independent failure signals, deliberately OR-ed: a file can report `not
# ok` lines while its process still exits 0 (a bats formatter or wrapper that
# swallows the status), and a file can exit non-zero having emitted no `not
# ok` at all (a crash, a timeout, an unbound variable in setup_file). Real
# bats normally emits both at once, so each signal masks the other and
# dropping either half is invisible without tests that produce one without
# the other -- tests/test-run-bats.sh has those.
if [[ "$file_not_ok" -gt 0 || "$status" != "0" ]]; then
# Three independent failure signals, deliberately OR-ed: a file can report
# `not ok` lines while its process still exits 0 (a bats formatter or wrapper
# that swallows the status), a file can exit non-zero having emitted no `not
# ok` at all (a crash, a timeout, an unbound variable in setup_file), and a
# file can emit FEWER results than its own plan line promised. Real bats
# normally emits all three consistently, so each signal masks the others and
# dropping any one of them is invisible without tests that produce one without
# the rest -- tests/test-run-bats.sh has those.
#
# The plan is the third signal and it is now enforced, not merely counted. It
# is the one that survives precisely the wrapper-swallows-the-status case
# named above: a process printing `1..10`, three `ok` lines and exit 0 used to
# be counted as "3 tests, 0 failures" and go green with seven tests silently
# gone, because the plan was computed for the aggregate zero-count guard below
# and then discarded. Mismatch either way is a failure -- more results than
# planned is as broken a TAP stream as fewer.
file_short=false
if [[ -n "$file_planned" && $((file_ok + file_not_ok)) -ne "$file_planned" ]]; then
file_short=true
echo "Error: $rel planned $file_planned test(s) but emitted $((file_ok + file_not_ok)) result line(s) — the run was truncated, or its exit status was swallowed" >&2
fi
if [[ "$file_not_ok" -gt 0 || "$status" != "0" || "$file_short" == true ]]; then
FAIL=1
fi
done

View File

@@ -41,6 +41,12 @@
# passing hook, and a SUGGESTION deliberately does not fail, so dropping
# one word from the config silences the tier ADR-0020 depends on while
# every test and every hook still reports green.
# 4. The WIRING of every repo-authored hook — its id, its `entry:` and its
# `stages:`. Same failure shape as 3, one level up: the tests drive these
# scripts by path, so nothing noticed whether .pre-commit-config.yaml
# still invoked them. On a scratch copy, deleting eight local hook blocks
# at once — `run-tests` among them — left every suite green. Upstream
# hooks and every `rev:` are deliberately out of scope.
set -euo pipefail
REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
@@ -680,6 +686,199 @@ while IFS=$'\t' read -r status msg; do
fi
done <<< "$VERBOSE_REPORT"
# ---------------------------------------------------------------------------
# 4. Every repo-authored hook is still WIRED, at the stage it claims
# ---------------------------------------------------------------------------
# Same failure mode as assertion 3, one level up. Every gate this repo owns is
# driven in tests by PATH -- tests/test-skill-version-bump.sh runs
# scripts/check-skill-version-bump.sh directly -- so the script staying correct
# and the hook still existing are independent facts, and only the first one was
# pinned. Deleting the `run-tests` block from .pre-commit-config.yaml (a routine
# merge-conflict casualty) stops the entire suite from running at pre-push
# forever, and `bash tests/run-tests.sh --strict` still prints all green: the
# suite cannot notice that nothing invokes it. Verified on a scratch copy --
# eight repo-authored hook blocks were deleted at once and every suite stayed
# green.
#
# So: id + entry + stages, for the repo-authored hooks only. "Repo-authored"
# is derived from the config (every hook under a `repo: local` entry), not
# hardcoded, and the derived set is then compared against the expected list
# below -- a new local hook that nobody pinned is itself a failure.
#
# Deliberately NOT pinned: the stock upstream hooks (gitleaks, check-yaml,
# pretty-format-json, shellcheck, conventional-pre-commit, the `repo: meta`
# pair) and every `rev:`. Those are somebody else's contract; a rev bump must
# not churn this test.
echo ""
echo "--- every repo-authored hook is wired, with the entry and stages it claims ---"
WIRING_REPORT="$(python3 - "$REPO_ROOT" <<'PY'
import os
import re
import sys
import yaml
root = sys.argv[1]
def emit(status, msg):
print("%s\t%s" % (status, msg))
# The expected wiring of every hook this repo authors: id -> (entry, stages).
# ADDING OR REMOVING A REPO-AUTHORED HOOK DELIBERATELY REQUIRES EDITING THIS
# LIST. That is the point -- the list is the second party to the agreement, so a
# hook cannot leave .pre-commit-config.yaml without someone saying so here.
# Upstream hooks are absent on purpose and must stay absent.
EXPECTED = {
# pre-push
'run-tests': (
'bash tests/run-tests.sh --strict', ['pre-push']),
'check-executables-allow-sync': (
'bash scripts/check-executables-allow-sync.sh', ['pre-push']),
'apm-audit-ci': (
'bash -c \'for d in . plugins/*/; do (cd "$d" && apm audit --ci) || '
'{ echo "apm audit --ci failed in $d" >&2; exit 1; }; done\'',
['pre-push']),
'check-apm-agents-valid': (
'bash scripts/check-apm-agents-valid.sh', ['pre-push']),
'apm-pack-check-clean': (
'apm pack --check-versions --check-clean --dry-run', ['pre-push']),
'check-scope-walkup-sync': (
'bash scripts/check-scope-walkup-sync.sh', ['pre-push']),
'check-skill-version-bump': (
'bash scripts/check-skill-version-bump.sh', ['pre-push']),
'validate-marketplace': (
'claude plugin validate --strict .claude-plugin/marketplace.json',
['pre-push']),
# pre-commit
'skill-size-check': (
'scripts/skill-size-check.sh', ['pre-commit']),
'check-rtk-prefix': (
'scripts/check-rtk-prefix.sh', ['pre-commit']),
'vale-audit-prefilter-skill': (
'plugins/kyberforge/.apm/skills/factory-audit/scripts/vale-wrap.sh',
['pre-commit']),
'vale-audit-prefilter-agent': (
'plugins/kyberforge/.apm/skills/factory-audit/scripts/vale-wrap.sh',
['pre-commit']),
}
path = os.path.join(root, '.pre-commit-config.yaml')
try:
with open(path, encoding='utf-8') as fh:
cfg = yaml.safe_load(fh) or {}
except Exception as exc:
emit('FAIL', '.pre-commit-config.yaml did not parse: %s' % exc)
sys.exit(0)
local = {}
duplicates = []
for repo in cfg.get('repos') or []:
if repo.get('repo') != 'local':
continue
for hook in (repo.get('hooks') or []):
hid = hook.get('id')
if hid in local:
duplicates.append(hid)
local[hid] = hook
# Non-vacuity, first. Every assertion below iterates the derived set, so an
# empty one would report nothing and the section would pass having checked
# nothing at all -- the same shape of silent hole this whole file exists to
# close.
if not EXPECTED:
emit('FAIL', 'the expected repo-authored hook list is empty — '
'every wiring assertion below is vacuous')
sys.exit(0)
if not local:
emit('FAIL', '.pre-commit-config.yaml declares no `repo: local` hooks at '
'all — every gate this repo authors has been unwired, or the '
'config moved and this assertion is now measuring nothing')
sys.exit(0)
emit('PASS', '.pre-commit-config.yaml declares %d repo-authored (`repo: local`) '
'hooks — the wiring assertions below are not vacuous' % len(local))
if duplicates:
emit('FAIL', '.pre-commit-config.yaml declares duplicate local hook ids '
'(%s) — pre-commit runs one of them and the other is dead '
'config' % ', '.join(sorted(set(duplicates))))
# The set, before the per-hook detail: a deleted block shows up here as a
# missing id even if nothing else in the file changed.
missing = sorted(set(EXPECTED) - set(local))
extra = sorted(set(local) - set(EXPECTED))
if missing or extra:
parts = []
if missing:
parts.append('NOT WIRED in .pre-commit-config.yaml: %s' % ', '.join(missing))
if extra:
parts.append('wired but not pinned in this test: %s' % ', '.join(extra))
emit('FAIL', 'the repo-authored hook set has drifted — %s. A hook that '
'leaves the config stops running at push time while every '
'test stays green; a hook that arrives unpinned can leave '
'again unnoticed. Fix the config, or update EXPECTED in '
'tests/test-adr0020-contract.sh deliberately.' % '; '.join(parts))
else:
emit('PASS', '.pre-commit-config.yaml wires exactly the %d expected '
'repo-authored hooks, no more and no fewer' % len(EXPECTED))
# Per hook: the entry that runs and the stage it runs at. Both are single
# tokens whose loss is invisible -- an entry repointed at a path that no longer
# exists makes pre-commit fail loudly, but an entry repointed at a DIFFERENT
# real script does not, and a hook moved off pre-push simply never fires.
entry_paths = []
for hid in sorted(EXPECTED):
hook = local.get(hid)
if hook is None:
continue # already reported as missing above
exp_entry, exp_stages = EXPECTED[hid]
got_entry = hook.get('entry')
got_stages = hook.get('stages')
problems = []
if got_entry != exp_entry:
problems.append('entry is %r, expected %r' % (got_entry, exp_entry))
if list(got_stages or []) != exp_stages:
problems.append('stages is %r, expected %r — a hook at the wrong stage '
'(or at none) never fires' % (got_stages, exp_stages))
if problems:
emit('FAIL', 'hook %s: %s' % (hid, '; '.join(problems)))
else:
emit('PASS', 'hook %s runs `%s` at %s'
% (hid, exp_entry, ','.join(exp_stages)))
# Collect the in-repo script/manifest paths the pinned entry names, so the
# pin cannot agree with a config that points at nothing.
for token in re.split(r'\s+', exp_entry):
token = token.strip('\'"')
if re.fullmatch(r'[A-Za-z0-9_.\-/]+\.(sh|json)', token):
entry_paths.append((hid, token))
# Every pinned entry path exists on disk. Without this, EXPECTED and the config
# could agree perfectly on a script that was deleted.
if not entry_paths:
emit('FAIL', 'no pinned entry named an in-repo script or manifest — the '
'existence check below iterated zero times and proved nothing')
else:
broken = ['%s -> %s' % (hid, token)
for hid, token in entry_paths
if not os.path.exists(os.path.join(root, token))]
if broken:
emit('FAIL', 'pinned hook entries point at files that do not exist: %s'
% ', '.join(broken))
else:
emit('PASS', 'all %d in-repo files named by a pinned hook entry exist '
'on disk' % len(entry_paths))
PY
)"
while IFS=$'\t' read -r status msg; do
[[ -n "$status" ]] || continue
if [[ "$status" == PASS ]]; then
pass "$msg"
else
fail "$msg"
fi
done <<< "$WIRING_REPORT"
echo ""
echo "Results: $PASS passed, $FAIL failed"
[[ $FAIL -eq 0 ]]

View File

@@ -204,7 +204,14 @@ while IFS= read -r rel; do
[[ -f "$REPO_ROOT/$rel" ]] && FILES+=("$REPO_ROOT/$rel")
done < <(git -C "$REPO_ROOT" ls-files -- '*.sh' '*.bats' '*.bash')
if [[ "${#FILES[@]}" -lt 20 ]]; then
# Floored at 50 against a real 57 (58 tracked `*.sh`/`*.bats`/`*.bash` files, less
# this one, which excludes itself above). Same discipline as the per-glob floors in
# tests/test-vale-wrap.sh: a guard against a broken `git ls-files` invocation or a
# moved search root resolving to a fraction of the tree, not a headcount to keep in
# step. Seven files of slack is one deliberate multi-file deletion -- a plugin's
# whole tests/ directory is three or four .bats files -- and nowhere near enough to
# absorb a discovery that degraded to a single directory.
if [[ "${#FILES[@]}" -lt 50 ]]; then
fail "only ${#FILES[@]} tracked shell files found — the scan is looking in the wrong place"
else
LIVE_HITS="$(scan ${FILES[@]+"${FILES[@]}"})"

View File

@@ -444,6 +444,38 @@ else
pass "a root under .claude/worktrees/ runs its own files and skips nested worktrees"
fi
# --- 11. A file that emits fewer results than its own plan promised fails the
# run. This is the third aggregation signal, and the only one left standing in
# exactly the case run-bats.sh's own comment puts in its threat model: a bats
# formatter or wrapper that swallows the exit status. A process printing `1..10`,
# three `ok` lines and exiting 0 emits no `not ok` and no non-zero status, so both
# other halves stay silent -- the plan was already being computed for the
# aggregate zero-count guard and was then thrown away, so the run was counted as
# "6 tests, 0 failures" and went green with fourteen tests silently gone.
echo ""
echo "--- a file emitting fewer results than its plan fails the run ---"
DIR11="$(make_fake_repo)"
FIXTURES+=("$DIR11")
seed_bats_files "$DIR11"
install_stub_bats "$DIR11" <<'EOF'
#!/usr/bin/env bash
echo "1..10"
echo "ok 1 first"
echo "ok 2 second"
echo "ok 3 third"
exit 0
EOF
run_fake "$DIR11"
if [[ $FAKE_RC -eq 0 ]]; then
fail "a file delivering 3 of its 10 planned tests exited 0 — a truncated run reported as a pass"
elif ! grep -q "planned 10 test(s) but emitted 3 result line(s)" <<< "$FAKE_OUT"; then
fail "the run failed but not with the plan-shortfall message: $FAKE_OUT"
elif grep -q "^6 tests, 0 failures$" <<< "$FAKE_OUT"; then
pass "a plan promising more tests than were delivered fails the run and names the shortfall"
else
fail "the plan-shortfall run failed with the wrong count: $FAKE_OUT"
fi
echo ""
echo "Results: $PASS passed, $FAIL failed"
[[ $FAIL -eq 0 ]]

View File

@@ -1,5 +1,5 @@
#!/usr/bin/env bash
# Regression test for scripts/vale-wrap.sh: Vale's `text.frontmatter.description`
# Regression test for plugins/kyberforge/.apm/skills/factory-audit/scripts/vale-wrap.sh: Vale's `text.frontmatter.description`
# NLP scope silently stops matching when the description value is a YAML block
# scalar spanning 2+ physical lines. vale-wrap.sh flattens it to one line before
# handing off to the real vale binary — this asserts that actually happens.