refactor!: carry out the simplification audit across gates, tests, plugins and docs #135

Merged
Defame1297 merged 85 commits from docs/simplification-audit into main 2026-09-20 19:14:03 +00:00
2 changed files with 293 additions and 53 deletions
Showing only changes of commit 1ce596cdbe - Show all commits

View File

@@ -6,42 +6,74 @@ set -euo pipefail
# is the gate that holds the rule, since skill-size-check only checks presence
# and shape.
#
# Baseline: `git merge-base <main> <pushed ref>`. Readers install skills from
# main, so "changed" means changed relative to what main ships, not relative to
# the remote branch's current tip. Diffing from PRE_COMMIT_FROM_REF would let
# the second push of a feature branch excuse a change the first push already
# carried unbumped. Unlike check-release-needed this runs on EVERY push, not
# only pushes to main: a missing bump is cheapest to fix on the branch, before
# review, and a push to main is then already covered.
# Baseline: `git merge-base <main> <pushed commit>`, where <main> is origin/main
# when it resolves and the local `main` branch otherwise. Readers install
# skills from main, so "changed" means changed relative to what main ships, not
# relative to the remote branch's current tip. Diffing from PRE_COMMIT_FROM_REF
# would let the second push of a feature branch excuse a change the first push
# already carried unbumped. The check runs on every push whatever the target
# branch — nothing here reads PRE_COMMIT_REMOTE_BRANCH — so it also runs under
# a manual `pre-commit run --hook-stage pre-push` (against HEAD, since no
# PRE_COMMIT_TO_REF is set). A missing bump is cheapest to fix on the branch,
# before review.
#
# Pushing main itself: with origin/main as the baseline, a push of main diffs
# the new commits against what the remote already has, so it is covered. A
# pushed commit that is already an ancestor of origin/main (merge-base equals
# the pushed commit) changes nothing relative to main and passes. With only the
# local `main` fallback, pushing main makes the merge-base the pushed commit
# itself — the diff is empty by construction, not because nothing changed — so
# that combination FAILS closed rather than passing unchecked.
#
# Scope: every skill directory plugins/<plugin>/.apm/skills/<skill>/, bin
# included. Anything under <skill>/tests/ is ignored — no agent ever loads it,
# so a test-only change ships nothing to a reader. A skill counts as changed
# when any other file under its directory differs between baseline and pushed
# ref. Renames are diffed as delete + add (--no-renames), so:
# commit. Paths are read NUL-delimited (`git diff -z`), so core.quotePath never
# hides a non-ASCII path. Renames are diffed as delete + add (--no-renames), so:
# - a skill absent at the baseline (new, renamed-to, merged-into) is exempt;
# it has no prior version to exceed.
# - a skill absent at the pushed ref (deleted, renamed-from) is exempt;
# there is nothing left to version.
# - a skill directory absent at the pushed commit (deleted, renamed-from) is
# exempt; there is nothing left to version. A directory that survives
# without its SKILL.md is NOT exempt: it fails as "SKILL.md missing".
# A changed skill present at both refs must carry a three-part semver
# `metadata.version` at the pushed ref (same SEMVER_RE as skill-size-check.sh,
# so pre-release suffixes are rejected here as they are there) that is
# numerically greater than the baseline's. A baseline with no parseable
# version (a skill predating ADR-0022) accepts any valid version. Versions are
# read from git objects, never the working tree.
# `metadata.version` at the pushed commit that is numerically greater than the
# baseline's. The shape rule follows skill-size-check.sh (str()-coerce, strip
# whitespace and quotes, three dot-separated numbers, so `1.0` and `1.0.0-rc1`
# are rejected) but is deliberately stricter: ASCII digits only (Python's `\d`
# also matches e.g. U+FF11) and at most 9 digits per part, so every part fits
# bash arithmetic. A baseline with no valid version (a skill predating
# ADR-0022) accepts any valid version. Versions are read from git objects,
# never the working tree.
#
# Missing baseline: if neither origin/main nor main resolves, or no merge-base
# exists (shallow clone, unrelated history), the gate FAILS closed — the same
# choice check-release-needed makes for an unreadable tag. Passing would make
# a fresh clone without main the one place the rule is silently off.
# Fails closed: if neither origin/main nor main resolves, if no merge-base
# exists (shallow clone, unrelated history), if the pushed ref does not resolve
# to a commit, if python3 or PyYAML is unavailable, or if a SKILL.md cannot be
# read. Passing in any of those would make that environment the one place the
# rule is silently off.
#
# Known gap: this only fires on a local `git push` through pre-commit's
# pre-push hook (or a manual `pre-commit run --hook-stage pre-push`). A PR
# merged via Gitea's merge button runs no local hook at all — closing that
# requires a server-side CI job, which this repo does not have yet.
# Known gaps:
# - Only the first pushed ref is gated. pre-commit (4.x, hook_impl.py
# `_pre_push_ns`) consumes the pre-push stdin itself and builds the
# environment from the first ref line that is not a delete and has commits
# the remote lacks; every later ref in the same `git push` (e.g.
# `git push origin a b`, `--all`, `--tags`) is never seen. When that first
# ref's unpushed history reaches a root commit, pre-commit runs with
# all_files and sets no PRE_COMMIT_TO_REF at all, so this script checks
# HEAD — which is the pushed ref only if it happens to be checked out. The
# script cannot recover either case: the ref list is gone by the time it
# runs. Push refs one at a time to be sure each is checked.
# - A PR merged via Gitea's merge button runs no local hook at all (the same
# gap check-release-needed has). Closing it requires a server-side CI job,
# which this repo does not have yet.
# See check-release-needed.sh: PRE_COMMIT_TO_REF is the local sha actually
# being pushed, which is only HEAD for the common case.
# Byte-wise regex matching and messages: path bytes are matched against
# SKILL_PATH_RE below and must not depend on the caller's locale.
export LC_ALL=C
# PRE_COMMIT_TO_REF is the local object actually being pushed, which is only
# HEAD for the common case. For a tag push it is the tag object, so it is
# peeled to a commit below before use.
PUSHED_REF="${PRE_COMMIT_TO_REF:-HEAD}"
# All-zeros sha: the push deletes a branch. Nothing ships; bail out.
@@ -52,6 +84,23 @@ fi
REPO_ROOT="$(git rev-parse --show-toplevel)"
cd "$REPO_ROOT"
if ! command -v python3 > /dev/null 2>&1; then
echo "FAIL: python3 is required to read SKILL.md metadata.version but was not found on PATH." >&2
echo " Fix: install python3 (pre-commit itself is a Python application, so it is almost certainly already present)." >&2
exit 1
fi
if ! python3 -c 'import yaml' > /dev/null 2>&1; then
echo "FAIL: PyYAML is required to read SKILL.md metadata.version but is not importable by python3." >&2
echo " Fix: python3 -m pip install PyYAML (or your distro's python3-yaml package)." >&2
exit 1
fi
if ! PUSHED_COMMIT="$(git rev-parse --verify -q "$PUSHED_REF^{commit}")"; then
echo "FAIL: pushed ref $PUSHED_REF does not resolve to a commit." >&2
exit 1
fi
MAIN_REF=""
for candidate in origin/main main; do
if git rev-parse --verify -q "$candidate^{commit}" > /dev/null; then
@@ -66,20 +115,29 @@ if [[ -z "$MAIN_REF" ]]; then
exit 1
fi
if ! BASELINE="$(git merge-base "$MAIN_REF" "$PUSHED_REF" 2>/dev/null)"; then
if ! BASELINE="$(git merge-base "$MAIN_REF" "$PUSHED_COMMIT" 2>/dev/null)"; then
echo "FAIL: no merge-base between $MAIN_REF and $PUSHED_REF, so there is no baseline to compare skill versions against." >&2
echo " Fix: ensure full history is available (e.g. git fetch --unshallow) and retry." >&2
exit 1
fi
if ! CHANGED="$(git diff --no-renames --name-only "$BASELINE" "$PUSHED_REF" -- plugins)"; then
if [[ "$MAIN_REF" == "main" && "$BASELINE" == "$PUSHED_COMMIT" ]]; then
echo "FAIL: origin/main does not resolve and $PUSHED_REF is already contained in local main, so local main cannot serve as an independent baseline — the diff would be empty by construction." >&2
echo " Fix: git fetch origin main and retry." >&2
exit 1
fi
CHANGED_FILE="$(mktemp)"
trap 'rm -f "$CHANGED_FILE"' EXIT
if ! git diff -z --no-renames --name-only "$BASELINE" "$PUSHED_COMMIT" -- plugins > "$CHANGED_FILE"; then
echo "FAIL: could not diff $BASELINE..$PUSHED_REF (see git error above)." >&2
exit 1
fi
SKILL_PATH_RE='^(plugins/[^/]+/\.apm/skills/[^/]+)/(.+)$'
SKILL_DIRS=()
while IFS= read -r path; do
while IFS= read -r -d '' path; do
[[ "$path" =~ $SKILL_PATH_RE ]] || continue
[[ "${BASH_REMATCH[2]}" == tests/* ]] && continue
dir="${BASH_REMATCH[1]}"
@@ -88,37 +146,54 @@ while IFS= read -r path; do
[[ "$existing" == "$dir" ]] && { seen=true; break; }
done
$seen || SKILL_DIRS+=("$dir")
done <<< "$CHANGED"
done < "$CHANGED_FILE"
[[ ${#SKILL_DIRS[@]} -eq 0 ]] && exit 0
# Prints the three-part semver metadata.version from SKILL.md on stdin, or
# nothing when it is missing or malformed. Same acceptance rule as
# skill-size-check.sh: str()-coerce, strip whitespace and quotes, then
# ^\d+\.\d+\.\d+$ — so `1.0` (a YAML float) and `1.0.0-rc1` both print nothing.
# Reads SKILL.md bytes on stdin and prints exactly one line: `OK <version>`
# when metadata.version is a valid three-part semver, `INVALID` when it is
# missing or malformed (including unparseable frontmatter). Any other outcome —
# python3 crashing, PyYAML failing to import — is a non-zero exit with no OK /
# INVALID line, which the caller reports as a read failure, never as a missing
# version. Bytes are decoded explicitly so the caller's locale cannot turn a
# non-ASCII SKILL.md into a crash; `\s*` before each `\n` absorbs CRLF.
read_version() {
python3 -c '
import re, sys, yaml
text = sys.stdin.read()
m = re.match(r"^---\s*\n(.*?)\n---\s*(\n|$)", text, re.S)
if not m:
sys.exit(0)
text = sys.stdin.buffer.read().decode("utf-8", errors="replace")
m = re.match(r"---[ \t\r]*\n(.*?)\n---[ \t\r]*(\n|\Z)", text, re.S)
data = None
if m:
try:
data = yaml.safe_load(m.group(1))
except Exception:
sys.exit(0)
except yaml.YAMLError:
data = None
meta = data.get("metadata") if isinstance(data, dict) else None
ver = meta.get("version") if isinstance(meta, dict) else None
if ver is None:
sys.exit(0)
ver = str(ver).strip().strip("\x27\"")
if re.match(r"^\d+\.\d+\.\d+$", ver):
print(ver)
ver = None if ver is None else str(ver).strip().strip("\x27\"")
if ver is not None and re.fullmatch(r"[0-9]{1,9}\.[0-9]{1,9}\.[0-9]{1,9}", ver):
print("OK " + ver)
else:
print("INVALID")
'
}
# Exit 0 when $1 > $2, both MAJOR.MINOR.PATCH, compared numerically so
# 1.0.10 > 1.0.9. 10# forces base 10 on a leading zero.
# version_at <commit> <path>: sets VERSION to the valid version or "" when
# invalid. Exits the script on a read failure.
version_at() {
local out
if ! out="$(git show "$1:$2" | read_version)" || [[ "$out" != OK\ * && "$out" != INVALID ]]; then
echo "FAIL: could not read metadata.version from $1:$2 (see error above)." >&2
exit 1
fi
VERSION=""
[[ "$out" == OK\ * ]] && VERSION="${out#OK }"
return 0
}
# Exit 0 when $1 > $2, both MAJOR.MINOR.PATCH with parts of at most 9 ASCII
# digits, compared numerically so 1.0.10 > 1.0.9. 10# forces base 10 on a
# leading zero.
semver_gt() {
local -a a b
local i
@@ -135,11 +210,18 @@ OFFENDERS=()
for dir in ${SKILL_DIRS[@]+"${SKILL_DIRS[@]}"}; do
# Absent at baseline: new, renamed-to, or merged-into. Exempt.
git cat-file -e "$BASELINE:$dir/SKILL.md" 2>/dev/null || continue
# Absent at pushed ref: deleted or renamed-from. Exempt.
[[ "$(git cat-file -t "$PUSHED_REF:$dir" 2>/dev/null)" == "tree" ]] || continue
# Directory absent at pushed commit: deleted or renamed-from. Exempt.
[[ "$(git cat-file -t "$PUSHED_COMMIT:$dir" 2>/dev/null)" == "tree" ]] || continue
base_ver="$(git show "$BASELINE:$dir/SKILL.md" | read_version)"
cur_ver="$(git show "$PUSHED_REF:$dir/SKILL.md" 2>/dev/null | read_version || true)"
version_at "$BASELINE" "$dir/SKILL.md"
base_ver="$VERSION"
if ! git cat-file -e "$PUSHED_COMMIT:$dir/SKILL.md" 2>/dev/null; then
OFFENDERS+=("$dir: SKILL.md missing at $PUSHED_REF (baseline: ${base_ver:-none})")
continue
fi
version_at "$PUSHED_COMMIT" "$dir/SKILL.md"
cur_ver="$VERSION"
if [[ -z "$cur_ver" ]]; then
OFFENDERS+=("$dir: metadata.version missing or not MAJOR.MINOR.PATCH at $PUSHED_REF (baseline: ${base_ver:-none})")

View File

@@ -188,9 +188,11 @@ F="$(make_fixture)"
write_skill "$F" demo alpha 'version: "1.0.0"' "unbumped"; commit "$F"
BAD="$(cd "$F" && git rev-parse HEAD)"
(cd "$F" && git checkout -q -b clean main)
echo "unrelated" > "$F/README.md"; commit "$F"
CLEAN="$(cd "$F" && git rev-parse HEAD)"
expect_fail "unbumped TO_REF fails while HEAD is clean" "alpha: 1\.0\.0 -> 1\.0\.0" "$F" "$BAD"
(cd "$F" && git checkout -q "$BAD")
expect_pass "clean TO_REF passes while HEAD is unbumped" "$F" "$(cd "$F" && git rev-parse main)"
expect_pass "clean TO_REF passes while HEAD is unbumped" "$F" "$CLEAN"
echo ""
echo "--- 14. all-zeros delete sha no-ops ---"
@@ -247,6 +249,162 @@ F="$(mktemp -d)"; CLEANUP_DIRS+=("$F")
write_skill "$F" demo alpha 'version: "1.0.0"'; commit "$F"
expect_fail "missing main fails with a clear message" "neither origin/main nor main resolves" "$F"
echo ""
echo "--- 20. non-ASCII paths are not hidden by core.quotePath ---"
F="$(make_fixture)"
mkdir -p "$F/plugins/demo/.apm/skills/alpha/references"
echo "ref" > "$F/plugins/demo/.apm/skills/alpha/references/résumé.md"; commit "$F"
expect_fail "non-ASCII file under references/ without bump fails" "alpha: 1\.0\.0 -> 1\.0\.0" "$F"
F="$(make_fixture)"
(cd "$F" && git checkout -q main)
write_skill "$F" demo "café" 'version: "2.0.0"'; commit "$F" "add café"
(cd "$F" && git checkout -q feature && git merge -q main)
write_skill "$F" demo "café" 'version: "2.0.0"' "changed"; commit "$F"
expect_fail "non-ASCII skill dir without bump fails" "skills/café: 2\.0\.0 -> 2\.0\.0" "$F"
write_skill "$F" demo "café" 'version: "2.0.1"' "changed again"; commit "$F"
expect_pass "non-ASCII skill dir with bump passes" "$F"
echo ""
echo "--- 21. local main fallback: pushed commit already in main fails closed ---"
F="$(make_fixture)"
MAIN_SHA="$(cd "$F" && git rev-parse main)"
expect_fail "pushing main's sha without origin/main fails" \
"origin/main does not resolve.*already contained in local main" "$F" "$MAIN_SHA"
(cd "$F" && git checkout -q main)
expect_fail "HEAD on main without origin/main fails" "Fix: git fetch origin main" "$F"
echo ""
echo "--- 22. pushing main itself with origin/main present ---"
F="$(make_fixture)"
(cd "$F" && git checkout -q main && git update-ref refs/remotes/origin/main main)
expect_pass "main equal to origin/main passes (nothing changed vs main)" "$F"
write_skill "$F" demo alpha 'version: "1.0.1"' "x"; commit "$F"
expect_pass "main ahead of origin/main with a bump passes" "$F"
write_skill "$F" demo beta "version: 1.0.9" "x"; commit "$F"
expect_fail "main ahead of origin/main without a bump fails" "beta: 1\.0\.9 -> 1\.0\.9" "$F"
(cd "$F" && git update-ref refs/remotes/origin/main main)
expect_pass "already-merged content (merge-base == pushed) passes against origin/main" "$F" \
"$(cd "$F" && git rev-parse main~1)"
echo ""
echo "--- 23. version shape is ASCII-only and bounded ---"
for v in 'version: "1.0.1"' 'version: "1.0.1"' 'version: "1.0.9999999999"' \
'version: "99999999999999999999.0.0"'; do
F="$(make_fixture)"
write_skill "$F" demo alpha "$v" "new body"; commit "$F"
OUT="$(run_check "$F" || true)"
if grep -q "alpha: metadata\.version missing or not" <<< "$OUT" \
&& ! grep -qiE "integer|syntax error|value too great" <<< "$OUT"; then
pass "'$v' is rejected as invalid without a bash arithmetic error"
else
fail "'$v' not cleanly rejected: $OUT"
fi
done
F="$(make_fixture)"
write_skill "$F" demo alpha 'version: "1.0.999999999"' "new body"; commit "$F"
expect_pass "nine-digit part is accepted and compared" "$F"
echo ""
echo "--- 24. python3 / PyYAML failures are never reported as a missing version ---"
REAL_PYTHON="$(command -v python3)"
SHIM_ROOT="$(mktemp -d)"; CLEANUP_DIRS+=("$SHIM_ROOT")
mkdir -p "$SHIM_ROOT/shadow" "$SHIM_ROOT/noyaml" "$SHIM_ROOT/crash"
printf 'raise ImportError("PyYAML deliberately unavailable in this fixture")\n' \
> "$SHIM_ROOT/shadow/yaml.py"
cat > "$SHIM_ROOT/noyaml/python3" <<EOF
#!/bin/sh
PYTHONPATH="$SHIM_ROOT/shadow\${PYTHONPATH:+:\$PYTHONPATH}" exec "$REAL_PYTHON" "\$@"
EOF
# Passes the up-front import probe, crashes on the real read.
cat > "$SHIM_ROOT/crash/python3" <<EOF
#!/bin/sh
[ "\$1" = "-c" ] && [ "\$2" = "import yaml" ] && exec "$REAL_PYTHON" "\$@"
echo "Traceback: simulated interpreter failure" >&2
exit 1
EOF
chmod +x "$SHIM_ROOT/noyaml/python3" "$SHIM_ROOT/crash/python3"
if PATH="$SHIM_ROOT/noyaml:$PATH" python3 -c 'import yaml' 2>/dev/null; then
fail "fixture check: the no-PyYAML shim still imports yaml — the next assertion would be vacuous"
else
pass "fixture check: the no-PyYAML shim makes 'import yaml' fail"
fi
F="$(make_fixture)"
write_skill "$F" demo alpha 'version: "1.0.1"' "new body"; commit "$F"
OUT="$(cd "$F" && unset PRE_COMMIT_FROM_REF PRE_COMMIT_TO_REF PRE_COMMIT_REMOTE_BRANCH \
&& PATH="$SHIM_ROOT/noyaml:$PATH" bash "$SCRIPT" 2>&1)" && RC=0 || RC=$?
if [[ $RC -ne 0 ]] && grep -q "PyYAML is required" <<< "$OUT" && grep -q "Fix: python3 -m pip install PyYAML" <<< "$OUT" \
&& ! grep -q "Traceback" <<< "$OUT"; then
pass "missing PyYAML fails with FAIL + Fix, no traceback"
else
fail "missing PyYAML not reported cleanly (rc=$RC): $OUT"
fi
OUT="$(cd "$F" && unset PRE_COMMIT_FROM_REF PRE_COMMIT_TO_REF PRE_COMMIT_REMOTE_BRANCH \
&& PATH="$SHIM_ROOT/crash:$PATH" bash "$SCRIPT" 2>&1)" && RC=0 || RC=$?
if [[ $RC -ne 0 ]] && grep -q "could not read metadata.version" <<< "$OUT" \
&& ! grep -q "missing or not" <<< "$OUT"; then
pass "python3 crash during the read is a read failure, not a missing version"
else
fail "python3 crash misreported (rc=$RC): $OUT"
fi
echo ""
echo "--- 25. SKILL.md deleted but skill dir kept ---"
F="$(make_fixture)"
mkdir -p "$F/plugins/demo/.apm/skills/alpha/references"
echo "ref" > "$F/plugins/demo/.apm/skills/alpha/references/x.md"
rm "$F/plugins/demo/.apm/skills/alpha/SKILL.md"; commit "$F"
expect_fail "missing SKILL.md is reported as such" "alpha: SKILL\.md missing at HEAD \(baseline: 1\.0\.0\)" "$F"
echo ""
echo "--- 26. baseline without a valid version accepts any valid version ---"
F="$(make_fixture)"
(cd "$F" && git checkout -q main)
write_skill "$F" demo legacy ""; commit "$F" "add legacy skill"
(cd "$F" && git checkout -q feature && git merge -q main)
write_skill "$F" demo legacy 'version: "0.0.1"' "changed"; commit "$F"
expect_pass "invalid baseline + valid current passes" "$F"
echo ""
echo "--- 27. no merge-base (unrelated histories) fails closed ---"
F="$(make_fixture)"
(cd "$F" && git checkout -q --orphan unrelated && git rm -rq --cached . && rm -rf plugins)
write_skill "$F" demo alpha 'version: "1.0.0"' "orphan"; commit "$F" "orphan root"
expect_fail "unrelated history fails at the merge-base check" "no merge-base between main and HEAD" "$F"
echo ""
echo "--- 28. annotated tag push is peeled to its commit ---"
F="$(make_fixture)"
write_skill "$F" demo alpha 'version: "1.0.0"' "unbumped"; commit "$F"
(cd "$F" && git tag -a v9 -m "tag" && git checkout -q main)
TAG_OBJ="$(cd "$F" && git rev-parse v9)"
if [[ "$(cd "$F" && git cat-file -t "$TAG_OBJ")" == "tag" ]]; then
pass "fixture check: TO_REF is a tag object, not a commit"
else
fail "fixture check: v9 is not an annotated tag object"
fi
expect_fail "unbumped change behind an annotated tag fails" "alpha: 1\.0\.0 -> 1\.0\.0" "$F" "$TAG_OBJ"
echo ""
echo "--- 29. CRLF frontmatter is parsed ---"
# crlf <repo> <skill>: rewrite that skill's SKILL.md with CRLF line endings.
crlf() {
local f="$1/plugins/demo/.apm/skills/$2/SKILL.md"
sed 's/$/\r/' "$f" > "$f.tmp" && mv "$f.tmp" "$f"
}
F="$(make_fixture)"
(cd "$F" && git config core.autocrlf false && git checkout -q main)
crlf "$F" alpha; commit "$F" "alpha to CRLF"
(cd "$F" && git checkout -q feature && git merge -q main)
write_skill "$F" demo alpha 'version: "1.0.0"' "crlf body"; crlf "$F" alpha; commit "$F"
if grep -q $'\r' "$F/plugins/demo/.apm/skills/alpha/SKILL.md"; then
pass "fixture check: SKILL.md carries CRLF"
else
fail "fixture check: SKILL.md has no CRLF"
fi
expect_fail "unbumped CRLF skill reports both parsed versions" "alpha: 1\.0\.0 -> 1\.0\.0" "$F"
write_skill "$F" demo alpha 'version: "1.0.1"' "crlf body 2"; crlf "$F" alpha; commit "$F"
expect_pass "bumped CRLF skill passes" "$F"
echo ""
echo "Results: $PASS passed, $FAIL failed"
[[ $FAIL -eq 0 ]]