Commit Graph
2 Commits
Author SHA1 Message Date
2973ca593d ci: pin every GitHub Action to a commit SHA (#4671)
## Description

Every third-party action in `.github/` ran from a version tag like
`actions/checkout@v7`. A tag can be moved to other code at any time.
That is how `tj-actions/changed-files` leaked CI secrets from ~23,000
repos in March 2025.

Here a moved tag would run with the release job's PyPI OIDC token
(`id-token: write`) and the Docker job's `packages: write` token.
Pinning to a commit SHA closes that, and a new test keeps it closed.

Closes #

## Type of Change

- [ ] Bug fix (non-breaking change that fixes an issue)
- [ ] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [ ] Documentation update
- [ ] Performance improvement
- [x] Code refactoring (no functional changes)

## Changes Made

- All 195 third-party `uses:` refs in 25 files now name the commit their
tag points to today, with the version as a comment:
`actions/checkout@3d3c42e5... # v7.0.1`. No action version changes.
- Dependabot's existing `github-actions` config updates SHA pins and
their version comments, so the pins stay current.
- Pin dtolnay/rust-toolchain stable and 1.95.0 branches to their
existing commit heads. The stable action retains its runtime toolchain
input/default; the 1.95.0 action embeds toolchain: 1.95.0. This
preserves each call site's existing toolchain selection.
- New `test_every_action_is_pinned_to_a_commit_sha` in
`scripts/tests/test_ci_workflow.py` fails on any tag-pinned action and
lists each one.
- Three tests in `tests/test_release_workflows.py` matched the old tag
text (`setup-buildx-action@v4`, `gh-action-pypi-publish@vX.Y.Z`,
`release-please-action@v4/v5`). They now match the SHA pin and its
version comment.

## Testing

- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [ ] Type checking passes (`mypy headroom`)
- [x] New tests added for new functionality
- [x] Manual testing performed

### Test Output

```text
$ uv run --frozen --extra dev pytest -q scripts/tests/test_ci_workflow.py scripts/tests/test_pr_health_workflow.py scripts/tests/test_normalize_release_pr_bodies.py tests/test_release_workflows.py
87 passed in 0.87s

$ uv run --frozen --extra dev pytest -q scripts/tests/test_pr_health_labels.py scripts/tests/test_version_sync.py tests/test_cli/test_wrap_copilot.py tests/test_learn/test_subagent_scanning.py tests/test_plugin_manifests.py tests/test_pr208_changes.py
114 passed in 1.32s

$ uvx ruff check scripts/tests/test_ci_workflow.py tests/test_release_workflows.py && uvx ruff format --check <same>
All checks passed!
```

## Real Behavior Proof

- Environment: macOS, Python 3.12 via `uv run --frozen --extra dev`,
branch cut from `main` at c01cf07d.
- Exact command / steps: resolved each tag with `git ls-remote --tags
--heads https://github.com/<owner>/<repo>` (peeled for annotated tags),
rewrote only the `uses:` lines, then ran the test files above and a
`yaml.safe_load` over every file in `.github/`.
- Observed result: 195 lines changed, 195 removed, no other line touched
(mixed LF/CRLF in `pr-health.yml` kept). All 201 tests pass. On `main`
the new test fails and lists the unpinned refs.
- Not tested: `scripts/validate-workflows.sh` (actionlint plus act dry
runs) was not run locally. The `workflow-validation` job runs it on this
PR.

## Runtime Rollout Safety

- Rollout-managed feature(s): None. CI workflow files and their tests
only.
- Minimum rollout channel: N/A, no runtime code changes.
- Stable/default behavior changed: No. Each SHA is the commit its tag
points to today.
- Kill switch / disable path: N/A, revert the commit.
- Unsafe override required: No.
- Qualification impact: None.
- Rollback path: Revert this PR.

## Review Readiness

- [x] I have performed a self-review
- [x] This PR is ready for human review

## Checklist

- [x] My code follows the project's style guidelines
- [x] I have performed a self-review of my code
- [x] I have commented my code, particularly in hard-to-understand areas
- [ ] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [x] New and existing unit tests pass locally with my changes
- [x] I did **not** edit `CHANGELOG.md` — it is generated by
release-please from my Conventional Commit PR title (a CI guard enforces
this)

## Screenshots (if applicable)

N/A, no UI change.

## Additional Notes

- mypy and docs are N/A: no `headroom/` code or documented behavior
changes.
- Dependabot does not bump branch pins, so the two
`dtolnay/rust-toolchain` pins stay on today's branch head until someone
moves them. That action rarely changes.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/headroomlabs-ai/headroom/pull/4671"
target="_blank"><picture><source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-devin-review-dark.svg?v=4"><img
src="https://static.devin.ai/assets/gh-devin-review-light.svg?v=4"
alt="Devin Review"></picture></a>
<!-- devin-review-badge-end -->

### Maintainer current-main requalification

Exact head `57b7edd8d05b6dc5e3bc4eac1dff55f21519afe8`, upstream main
`c01cf07d943d816cfd9d97e6838ed4e38f1ff9a4`.

Reviewed the27-file delta and existing discussion/Devin no-issues result
against current upstream main c01cf07d94,
already included in the contributor head. Independent full-workflow
comparison proves only uses refs/comments change across the25
workflow/action files. All195 replaced refs resolve exactly to their
original upstream tag/branch commits (29 unique pins across26
repositories), with annotated tags peeled; no mismatches. Pinned dtolnay
action source independently confirms stable runtime default versus
embedded1.95.0 branch behavior, preserving toolchain selection.87
workflow/release tests and113 adjacent tests pass,1skip; changed Python
lint/format passes. Maintainer fix only removes CRLF endings on two
edited pr-health lines to clear whitespace; provenance reverified
afterward. No contributor workflow logic altered. Description is
corrected to explain embedded1.95.0 semantics. No publish/release
workflow execution claimed; full hosted qualification must run on this
new mechanical-fix head. Maintainers handle CI. Nothing merged.


Exact-head hosted checks are being handled by maintainers. No new
finding in the reviewed scope; formal approval awaits exact-head hosted
qualification; nothing merged.

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: JerrettDavis <mxjerrett@gmail.com>
2026-11-06 09:16:15 -06:00
5698e91b2a chore(security): beacon rate limit, dev environment, weekly CodeQL triage (#4008)
SOC 2 follow-ups from the 2026-10-06 firewall review (R3, R4) and
incident INC-2026-001.

## Beacon rate limit (R3)
`deploy/beacon/wrangler.toml` said a WAF rule limited `/v1/logs` to 60
req/min/IP. That rule could not exist: headroomlabs.ai is not a
Cloudflare zone, so the Worker is only on `workers.dev`. This adds a
[Workers rate-limit
binding](https://developers.cloudflare.com/workers/runtime-apis/bindings/rate-limit/)
checked in `fetch()` **before** the body is read.

- Keyed on one constant, not the IP: the Worker still never reads
`cf-connecting-ip` (privacy promise in the file header). Trade-off: a
flood also sheds real events for that minute.
- 3000 req/min **per Cloudflare location** (~10x today's ~300/min
total).
- Returns 429, which the client does **not** treat as a gzip refusal
(`_GZIP_REFUSED_STATUSES` is 400/415), so rate-limited clients keep
compressing.
- Fails open if the binding is absent **or the limiter itself errors**
(review fix): a rate-limit service outage lets beacons through rather
than dropping all of them. `node deploy/beacon/test-rollup.mjs` covers
429 over the limit, fail-open on outage and when unbound, and the single
shared key.

## Dev environment (R4)
`[env.dev]` deploys `headroom-beacon-dev` bound to the new
`headroom-telemetry-dev` bucket (600/min limit). Already deployed and
tested: a sample event returns 204, and a 2,500-request burst got 384 ×
429.

## Weekly CodeQL triage
`.github/workflows/codeql-triage.yml` opens an issue every Monday,
assigned to @chopratejas (tejas@headroomlabs.ai), listing open
critical/high alerts with checkboxes. Alert #63 sat unowned for five
months; this gives alerts an owner and leaves an audit trail.

**One issue per week, however often it runs (review fix).** The logic is
in `scripts/codeql_triage.sh`:
- It keys the issue on the ISO week, using a hidden marker in the body.
- It finds the existing issue by listing security-labelled issues,
closed ones included. It uses a listing, not search, so an issue created
seconds earlier is not missed.
- It reuses that issue and never edits, reopens or retitles it, so
ticked boxes and closure stand. Alerts opened since then are added once,
as a comment.
- The jq programs live in quoted heredocs, which clears ShellCheck
SC2016 without any shell expansion.
- `tests/test_codeql_triage_script.py` runs the real script against a
fake `gh`. It covers same-day re-runs, a Friday dispatch on a closed
issue with ticked boxes, the next week, and the ISO year boundary
(2027-01-01 is in 2026-W53).

### Test output

```text
$ pytest -q tests/test_codeql_triage_script.py
6 passed
$ node deploy/beacon/test-rollup.mjs
rate limit: 429 over, fail-open on outage and when unbound
ok
$ shellcheck scripts/codeql_triage.sh && actionlint .github/workflows/codeql-triage.yml
(clean)
```

## ⚠️ Before deploying production
The live `headroom-beacon` Worker predates #3253 (gzip inflate, schema
v2 allowlist). `wrangler deploy` from main will ship that too, not just
this change.

🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/headroomlabs-ai/headroom/pull/4008"
target="_blank"><picture><source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-devin-review-dark.svg?v=4"><img
src="https://static.devin.ai/assets/gh-devin-review-light.svg?v=4"
alt="Devin Review"></picture></a>
<!-- devin-review-badge-end -->

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: JerrettDavis <mxjerrett@gmail.com>
2026-10-12 11:18:47 -07:00