Files
ocp/.github/workflows/alignment.yml
T
taodengandClaude Opus 5 66096cdcab ci: cover models.json in the alignment guardrail; fix release.yml's no-CHANGELOG path
Two CI-layer fixes found during the v3.25.0 review. Same layer, both small, so
they land together (Iron Rule 11).

#198 — alignment.yml did not run on a models.json-only PR.
The `paths:` filter listed server.mjs / setup.mjs / scripts / lib / ocp /
ocp-connect, but not models.json. That made CLAUDE.md hard-requirement #2 ("CI
blacklist pass") vacuous for exactly the PRs that change model routing:
confirmed on #192, where only gitleaks and test-features ran.

models.json is not inert data. It drives MODEL_MAP and VALID_MODELS, the
default request model, and — since ADR 0009 — the global MAX_PROMPT_CHARS
truncation budget. A models.json-only PR can therefore change server.mjs's
runtime behavior with the guardrail never running. models.schema.json is
included for the same reason one level up: loosening the schema is what would
let a malformed models.json through.

Adding paths only widens coverage; the blacklist grep still scans server.mjs,
so this costs nothing and closes the gap. Per CLAUDE.md this is a PR amendment
to alignment.yml, which is the sanctioned way to change it (removing entries
would need an ALIGNMENT.md amendment; nothing is removed here).

#202 — release.yml's no-CHANGELOG fallback would have failed the release job.
The branch wrote an output named `notes` and exited WITHOUT creating
/tmp/release-notes.md, while the create step hard-codes
`--notes-file /tmp/release-notes.md`. So the one path that exists to "degrade
to minimal notes" instead produced a failed release.

Verified both directions by extracting the step's actual script from the YAML
and running it, rather than reading it:

  PRE-FIX,  no CHANGELOG -> exit 0, GITHUB_OUTPUT="notes=Release v3.25.0",
                            /tmp/release-notes.md MISSING
                            -> gh release create --notes-file WOULD FAIL
  POST-FIX, no CHANGELOG -> exit 0, notes_file set, file present ("Release v3.25.0")
  POST-FIX, with CHANGELOG -> file present, first line "## v3.25.0 — 2026-07-27"

Fixed by writing the file in that branch, and by removing the hard-coded-path
coupling entirely: both branches now set `notes_file` and the create step
consumes `steps.notes.outputs.notes_file`. That output existed already and was
never read — the two only agreed by accident.

Also added `set -euo pipefail` (the step previously ran unset-tolerant) and an
echo of the resolved notes before creating the release, so a wrong-looking body
is visible in the job log instead of only on the published release.

NOT changed: the empty-extraction guard. My issue text suggested adding one,
but `if [ ! -s "$NOTES" ]` was already there and already handles a heading
mismatch. Correcting that claim on #202 rather than taking credit for it.

Both workflows re-validated as YAML.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017gbqUZ8HfBZpjjbzQ85oH8
2026-07-27 08:51:09 +10:00

215 lines
8.6 KiB
YAML

name: Alignment Guardrail
on:
pull_request:
paths:
- 'server.mjs'
- 'setup.mjs'
- 'scripts/**'
- 'lib/**'
- 'ocp'
- 'ocp-connect'
# models.json is NOT inert data: it drives MODEL_MAP and VALID_MODELS, the default
# request model, and — since ADR 0009 — the global MAX_PROMPT_CHARS truncation budget.
# A models.json-only PR can therefore change server.mjs's runtime behavior, and before
# this the guardrail simply did not run on one (#198). models.schema.json is included
# for the same reason one level up: loosening the schema is what would let a bad
# models.json through.
- 'models.json'
- 'models.schema.json'
- '.github/workflows/alignment.yml'
jobs:
blacklist:
name: server.mjs blacklist (hard fail)
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v4
- name: Scan server.mjs for hallucinated tokens
shell: bash
run: |
set -euo pipefail
if [ ! -f server.mjs ]; then
echo "server.mjs not found; nothing to scan."
exit 0
fi
# Blacklisted tokens — two kinds (see ALIGNMENT.md "OAuth token-host verification"):
# (1) known LLM hallucinations (e.g. the 2026-04-11 /api/oauth/usage drift), and
# (2) pinned wrong-host variants of a VERIFIED Class A endpoint (a hit means a
# drift to a known-wrong host, not necessarily a hallucination).
# Extend only via an ALIGNMENT.md amendment PR. Matched as fixed strings vs server.mjs.
BLACKLIST=(
"api.anthropic.com/api/oauth/usage"
"console.anthropic.com/v1/oauth/token"
)
FAIL=0
for token in "${BLACKLIST[@]}"; do
if sed "s|//.*\$||" server.mjs | grep -n -F "$token"; then
echo "::error file=server.mjs::Blacklisted token '$token' detected in server.mjs."
FAIL=1
fi
done
if [ "$FAIL" -ne 0 ]; then
cat <<'EOF'
============================================================
ALIGNMENT GUARDRAIL FAILURE
============================================================
server.mjs contains a token on the OCP alignment blacklist.
These tokens are either LLM hallucinations that never appeared in cli.js,
or pinned wrong-host variants of a verified Class A endpoint (a drift).
See ALIGNMENT.md -> "Historical Lesson: The 2026-04-11 Drift"
(commit b87992f) for the full incident record.
Required action:
1. Remove the token from server.mjs.
2. grep the reference cli.js for the operation you
intended and cite the real line numbers.
3. See ALIGNMENT.md Rules 1, 2, and 5.
Do not add allowlist entries to this workflow without an
amendment PR to ALIGNMENT.md (see Amendment Procedure).
============================================================
EOF
exit 1
fi
echo "Blacklist scan clean."
port-spot:
name: port literal SPOT (hard fail)
# Background: from 2026-05-08 (PR #71 dogfood accident) through 2026-05-13
# a hardcoded "3478" in scripts/upgrade.mjs + scripts/doctor.mjs cascaded
# into wrong baseUrl writes for the OpenClaw "claude-local" provider,
# taking out the "大内总管" Telegram agent.
#
# Rule: the only places allowed to write a literal port number in source
# are (a) lib/constants.mjs (the SPOT), (b) bash scripts ocp / ocp-connect
# (which can't import .mjs and must keep the literal in sync — flagged
# with a `// keep in sync with lib/constants.mjs` style comment), and
# (c) test-features.mjs (intentionally pins historical ports for plist /
# systemd parser tests). Everything else MUST import from lib/constants.mjs.
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v4
- name: Scan for hardcoded port literals outside SPOT
shell: bash
run: |
set -euo pipefail
# Files/paths exempt from the SPOT requirement.
EXEMPT_REGEX='^(lib/constants\.mjs|test-features\.mjs|ocp|ocp-connect|CHANGELOG\.md|README\.md|docs/|\.github/workflows/alignment\.yml)'
# Hardcoded port literals to forbid in non-exempt source.
FORBIDDEN_PORTS=("3478" "3456")
FAIL=0
for port in "${FORBIDDEN_PORTS[@]}"; do
HITS="$(git ls-files | grep -E '\.(mjs|js|ts|json)$' \
| xargs grep -n -E "[^0-9]${port}[^0-9]" 2>/dev/null \
| grep -v -E "${EXEMPT_REGEX}" \
|| true)"
if [ -n "$HITS" ]; then
echo "::error::Hardcoded port literal '${port}' found outside lib/constants.mjs:"
echo "$HITS"
FAIL=1
fi
done
if [ "$FAIL" -ne 0 ]; then
cat <<'EOF'
============================================================
PORT LITERAL SPOT VIOLATION
============================================================
A hardcoded TCP port literal was found in a source file
that should import from lib/constants.mjs instead.
Background: this rule exists because between 2026-05-08 and
2026-05-13 a stray hardcoded "3478" in scripts/upgrade.mjs
and scripts/doctor.mjs cascaded into downstream OpenClaw
config writes, taking out the OpenClaw Telegram agent.
See v3.16.3 CHANGELOG and lib/constants.mjs header comment.
Required action:
1. Import DEFAULT_PORT (or related constant) from
lib/constants.mjs instead of hardcoding the literal.
2. If the file genuinely cannot import .mjs (e.g. bash
script), add it to EXEMPT_REGEX in this workflow and
add a `keep in sync with lib/constants.mjs` comment
at the reference.
3. For test files that intentionally pin historical ports
(test-features.mjs), the regex already exempts them.
============================================================
EOF
exit 1
fi
echo "Port SPOT scan clean."
commit-citation:
name: commit message citation (soft check)
runs-on: ubuntu-latest
# Soft check: reports a warning but does not block merge. Reviewers
# are expected to enforce per CLAUDE.md. Escalate to hard-fail via
# an ALIGNMENT.md amendment if drift recurs.
continue-on-error: true
steps:
- name: Checkout full history
uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Scan PR commits for uncited assertions
shell: bash
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
HEAD_SHA: ${{ github.event.pull_request.head.sha }}
run: |
set -euo pipefail
if [ -z "${BASE_SHA:-}" ] || [ -z "${HEAD_SHA:-}" ]; then
echo "No PR context; skipping."
exit 0
fi
# Collect commit messages in range.
MSGS="$(git log --format=%B "${BASE_SHA}..${HEAD_SHA}")"
# Look for assertions of the form "Claude Code uses ..." or
# "cli.js uses ..." (case-insensitive). For each hit, require
# a citation in the same commit message in one of the forms:
# cli.js:NNNN (line-number citation)
# cli.js vE4 <name> (version + function-name citation)
# Absence of a citation is a soft finding.
WARN=0
# Split log into per-commit blocks for precise matching.
git log --format="%H" "${BASE_SHA}..${HEAD_SHA}" | while read -r sha; do
BODY="$(git log -1 --format=%B "$sha")"
if echo "$BODY" | grep -E -i -q '(claude[[:space:]]+code|cli\.js)[[:space:]]+uses'; then
if echo "$BODY" | grep -E -q '(cli\.js:[0-9]+|cli\.js[[:space:]]+v[A-Za-z0-9]+[[:space:]]+[A-Za-z_][A-Za-z0-9_]*)'; then
echo "OK $sha: assertion cited."
else
echo "::warning::Commit $sha asserts 'Claude Code uses ...' or 'cli.js uses ...' but does not cite a cli.js line number or versioned function name. See CLAUDE.md -> Commit message conventions."
WARN=1
fi
fi
done
if [ "$WARN" -ne 0 ]; then
echo "Soft check raised warnings. Reviewer: please enforce per CLAUDE.md."
else
echo "Commit citation soft check clean."
fi