Move release/announce into Deploy workflow to clear CodeQL

The workflow_run-triggered release.yaml kept tripping CodeQL's
privileged-context rules (untrusted-checkout, then cache-poisoning) —
CodeQL distrusts any workflow_run that checks out a ref, regardless of
the main-only guarantee, and there is no autofix.

Fold the release job into deploy.yaml gated on `needs: deploy`. A
push-to-main run is a trusted context, so checking out and running repo
code with write/OIDC is safe there. This still gates on deploy success
and serializes via the deploy concurrency group, and removes the
separate workflow entirely.
This commit is contained in:
Adam Moussa 2026-06-11 19:38:53 -04:00
parent 7e4458724f
commit ed32541331
3 changed files with 112 additions and 159 deletions

View file

@ -20,3 +20,101 @@ jobs:
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
parameter-overrides: ${{ secrets.SAM_PARAMETER_OVERRIDES }}
# Tag + announce a release once the deploy succeeds. This lives in the deploy
# workflow (gated on `needs: deploy`) rather than a separate workflow_run-
# triggered job on purpose: a push-to-main run is a trusted context, so
# checking out and running repo code with write/OIDC is safe here — unlike
# workflow_run, which CodeQL (rightly) flags for untrusted checkout + cache
# poisoning. Gating on `needs: deploy` still guarantees we never announce a
# version that isn't live, and the `deploy` concurrency group serializes
# releases. When the top CHANGELOG version already has a Release, this no-ops.
release:
needs: deploy
runs-on: ubuntu-latest
permissions:
contents: write # create the tag + GitHub Release
id-token: write # OIDC to assume the notifier-invoke role
steps:
- uses: actions/checkout@v6
with:
fetch-depth: 0
fetch-tags: true
- uses: actions/setup-python@v5
with:
python-version: "3.12"
- name: Determine release
id: rel
env:
GH_TOKEN: ${{ github.token }}
run: |
TOP=$(python scripts/changelog_cli.py top-version CHANGELOG.md)
if [ -z "$TOP" ]; then
echo "No version entry in CHANGELOG.md — nothing to release."
echo "release=false" >> "$GITHUB_OUTPUT"; exit 0
fi
PREV=$(git tag -l 'v*' --sort=-v:refname | head -1)
PREV="${PREV:-v0.0.0}"
KIND=$(python scripts/changelog_cli.py bump-kind CHANGELOG.md "$PREV")
RELEASE_EXISTS=false
gh release view "v$TOP" >/dev/null 2>&1 && RELEASE_EXISTS=true
echo "version=$TOP" >> "$GITHUB_OUTPUT"
echo "kind=$KIND" >> "$GITHUB_OUTPUT"
# Act only on a clean SemVer bump whose Release isn't published yet.
if [ "$KIND" != "none" ] && [ "$RELEASE_EXISTS" = "false" ]; then
echo "release=true" >> "$GITHUB_OUTPUT"
else
echo "release=false" >> "$GITHUB_OUTPUT"
echo "v$TOP: kind=$KIND release_exists=$RELEASE_EXISTS — no action."
fi
- name: Build release notes
if: ${{ steps.rel.outputs.release == 'true' }}
run: |
python scripts/changelog_cli.py payload CHANGELOG.md "${{ steps.rel.outputs.version }}" > payload.json
python -c "import json; print(json.load(open('payload.json'))['notes'])" > notes.md
# Announce BEFORE publishing the Release: the Release is the durable "done"
# marker (the step above skips once it exists), so announcing first keeps
# this retryable. Minor/major only, and only once the invoke-role variable
# has been bootstrapped (see README).
- name: Configure AWS credentials
if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
uses: aws-actions/configure-aws-credentials@v6
with:
role-to-assume: ${{ vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN }}
aws-region: us-east-1
- name: Announce in Slack
if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
run: |
aws lambda invoke \
--function-name afterhours-release-notifier \
--cli-binary-format raw-in-base64-out \
--payload file://payload.json \
--output json response.json > invoke-meta.json
# aws lambda invoke only emits a FunctionError key when the handler errored.
if grep -q '"FunctionError"' invoke-meta.json; then
echo "::error::release-notifier returned an error"; cat response.json; exit 1
fi
echo "Announced v${{ steps.rel.outputs.version }}."
- name: Warn if announcement skipped (not bootstrapped)
if: ${{ steps.rel.outputs.release == 'true' && steps.rel.outputs.kind != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN == '' }}
run: echo "::warning::RELEASE_NOTIFY_INVOKE_ROLE_ARN is unset — tagging + releasing but not announcing. Set the repo variable from the stack output."
- name: Publish GitHub Release
if: ${{ steps.rel.outputs.release == 'true' }}
env:
GH_TOKEN: ${{ github.token }}
run: |
# gh creates the tag at the deployed commit and the Release together.
gh release create "v${{ steps.rel.outputs.version }}" \
--repo "${{ github.repository }}" \
--title "v${{ steps.rel.outputs.version }}" \
--notes-file notes.md \
--target "${{ github.sha }}"

View file

@ -1,146 +0,0 @@
name: Release
# Runs after a successful Deploy. When the top of CHANGELOG.md names a version
# that has no Release yet, this tags it, publishes a GitHub Release with the
# notes, and — for minor/major bumps only — invokes the release-notifier Lambda
# to announce it in Slack. Triggering on Deploy completion (not release:published
# / tag push) is deliberate: GITHUB_TOKEN-created events do not start downstream
# workflows, and gating on Deploy success means we never announce a version that
# is not actually live.
#
# Two jobs by design (CodeQL actions/untrusted-checkout): the only job that
# checks out and runs repo code (`prepare`) is read-only and unprivileged; the
# job that holds write + OIDC (`publish`) never checks out repo code — it acts
# purely through the GitHub and AWS APIs.
on:
workflow_run:
workflows: ["Deploy"]
types: [completed]
# Serialize so back-to-back releases announce in order, never overlapping.
concurrency:
group: release-announce
cancel-in-progress: false
jobs:
prepare:
# Deploy only runs on push to main, so head_sha is always a trusted main
# commit; assert head_branch == main to make that boundary explicit.
if: ${{ github.event.workflow_run.conclusion == 'success' && github.event.workflow_run.head_branch == 'main' }}
runs-on: ubuntu-latest
permissions:
contents: read
outputs:
release: ${{ steps.rel.outputs.release }}
version: ${{ steps.rel.outputs.version }}
kind: ${{ steps.rel.outputs.kind }}
steps:
- uses: actions/checkout@v6
with:
ref: ${{ github.event.workflow_run.head_sha }}
fetch-depth: 0
fetch-tags: true
- uses: actions/setup-python@v5
with:
python-version: "3.12"
- name: Determine release
id: rel
env:
GH_TOKEN: ${{ github.token }}
run: |
TOP=$(python scripts/changelog_cli.py top-version CHANGELOG.md)
if [ -z "$TOP" ]; then
echo "No version entry in CHANGELOG.md — nothing to release."
echo "release=false" >> "$GITHUB_OUTPUT"
exit 0
fi
PREV=$(git tag -l 'v*' --sort=-v:refname | head -1)
PREV="${PREV:-v0.0.0}"
KIND=$(python scripts/changelog_cli.py bump-kind CHANGELOG.md "$PREV")
RELEASE_EXISTS=false
gh release view "v$TOP" >/dev/null 2>&1 && RELEASE_EXISTS=true
echo "version=$TOP" >> "$GITHUB_OUTPUT"
echo "kind=$KIND" >> "$GITHUB_OUTPUT"
# Gate the publish job on a clean bump whose Release isn't published yet.
if [ "$KIND" != "none" ] && [ "$RELEASE_EXISTS" = "false" ]; then
echo "release=true" >> "$GITHUB_OUTPUT"
else
echo "release=false" >> "$GITHUB_OUTPUT"
echo "v$TOP: kind=$KIND release_exists=$RELEASE_EXISTS — no action."
fi
- name: Build release notes
if: ${{ steps.rel.outputs.release == 'true' }}
run: |
python scripts/changelog_cli.py payload CHANGELOG.md "${{ steps.rel.outputs.version }}" > payload.json
python -c "import json; print(json.load(open('payload.json'))['notes'])" > notes.md
- name: Upload notes artifact
if: ${{ steps.rel.outputs.release == 'true' }}
uses: actions/upload-artifact@v4
with:
name: release-notes
path: |
payload.json
notes.md
retention-days: 1
publish:
needs: prepare
if: ${{ needs.prepare.outputs.release == 'true' }}
runs-on: ubuntu-latest
permissions:
contents: write # create the tag + GitHub Release
id-token: write # OIDC to assume the notifier-invoke role
env:
VERSION: ${{ needs.prepare.outputs.version }}
KIND: ${{ needs.prepare.outputs.kind }}
HEAD_SHA: ${{ github.event.workflow_run.head_sha }}
steps:
- uses: actions/download-artifact@v4
with:
name: release-notes
# Announce BEFORE publishing the Release: the Release is the durable "done"
# marker (prepare skips once it exists), so announcing first keeps this
# retryable. Minor/major only, and only once the invoke-role variable has
# been bootstrapped (see README).
- name: Configure AWS credentials
if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
uses: aws-actions/configure-aws-credentials@v6
with:
role-to-assume: ${{ vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN }}
aws-region: us-east-1
- name: Announce in Slack
if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN != '' }}
run: |
aws lambda invoke \
--function-name afterhours-release-notifier \
--cli-binary-format raw-in-base64-out \
--payload file://payload.json \
--output json response.json > invoke-meta.json
# aws lambda invoke only emits a FunctionError key when the handler errored.
if grep -q '"FunctionError"' invoke-meta.json; then
echo "::error::release-notifier returned an error"; cat response.json; exit 1
fi
echo "Announced v${VERSION}."
- name: Warn if announcement skipped (not bootstrapped)
if: ${{ env.KIND != 'patch' && vars.RELEASE_NOTIFY_INVOKE_ROLE_ARN == '' }}
run: echo "::warning::RELEASE_NOTIFY_INVOKE_ROLE_ARN is unset — tagging + releasing but not announcing. Set the repo variable from the stack output."
# No checkout: gh creates the tag at HEAD_SHA and the Release together.
- name: Publish GitHub Release
env:
GH_TOKEN: ${{ github.token }}
run: |
gh release create "v${VERSION}" \
--repo "${{ github.repository }}" \
--title "v${VERSION}" \
--notes-file notes.md \
--target "${HEAD_SHA}"

View file

@ -64,7 +64,7 @@ Dates accept: `today`, `tomorrow`, `monday`-`sunday`, `4/5`, `2026-04-05`
| `afterhours-weekly-post` | EventBridge (Monday 7am ET) | Posts weekly schedule to Slack, sends pay report email |
| `afterhours-roster-sync` | EventBridge (daily 6am ET) | Syncs employee roster from 3CX |
| `afterhours-ring-scheduler` | EventBridge (daily 8am ET + weekend 5pm ET) | Updates 3CX queue routing based on who's on shift |
| `afterhours-release-notifier` | Invoked by `release.yaml` on minor/major releases | Posts a "What's New" announcement to the shift channel |
| `afterhours-release-notifier` | Invoked by the Deploy workflow's release job on minor/major releases | Posts a "What's New" announcement to the shift channel |
### Project Layout
@ -131,18 +131,19 @@ bumping per SemVer, then run `python scripts/sync_changelog.py` to update the
in-package copy. The `Changelog Guard` PR check enforces that the bump is a clean
single SemVer step above the latest tag and that the two copies match.
On the **deploy-then-merge** path, once the merge's Deploy succeeds,
`.github/workflows/release.yaml` (triggered on Deploy completion) tags the new
version, publishes a GitHub Release with the notes, and — for **minor and major**
bumps only (patches stay silent) — invokes `afterhours-release-notifier` to post a
"What's New" message in the shift channel. The **App Home** tab ("About" page on
the bot) always shows the current version's notes, read from the CHANGELOG that
ships in the slack-bot package.
On the **deploy-then-merge** path, once the merge's Deploy succeeds, the Deploy
workflow's `release` job (`needs: deploy`) tags the new version, publishes a
GitHub Release with the notes, and — for **minor and major** bumps only (patches
stay silent) — invokes `afterhours-release-notifier` to post a "What's New"
message in the shift channel. The **App Home** tab ("About" page on the bot)
always shows the current version's notes, read from the CHANGELOG that ships in
the slack-bot package.
> Triggering on Deploy success (rather than `release: published` or a tag push) is
> deliberate: GitHub does not start downstream workflows from `GITHUB_TOKEN`-created
> events, and gating on a successful deploy guarantees we never announce a version
> that isn't actually live.
> The release job lives inside the Deploy workflow (gated on `needs: deploy`)
> rather than a separate `workflow_run`-triggered workflow. A push-to-main run is
> a trusted context, so checking out and running repo code with write/OIDC is safe
> — whereas `workflow_run` is flagged by CodeQL for untrusted checkout. Gating on
> `needs: deploy` still guarantees we never announce a version that isn't live.
**One-time setup (per environment):** after the first deploy creates the
`ReleaseNotifyInvokeRole`, copy its ARN from the `ReleaseNotifyInvokeRoleArn` stack
@ -153,7 +154,7 @@ and publish but skip the Slack announcement (with a warning).
> **Convention note (deliberate deviation).** The Sea Haven handbook says internal
> SAM stacks generally need no versioning and that tags are applied manually. This
> bot is versioned by owner choice (it has staff-facing release notes) and tagged
> automatically by `release.yaml`. This is intentional — not drift.
> automatically by the Deploy workflow's release job. This is intentional — not drift.
## Testing