Skip to content

docs(release): stop cutting Glama releases for the MCP server - #248

Merged
stewartjarod merged 1 commit into
mainfrom
advisor/release-drop-glama
Nov 3, 2026
Merged

stewartjarod merged 1 commit into
mainfrom
advisor/release-drop-glama

Conversation

@stewartjarod

Copy link
Copy Markdown
Contributor

Summary

/release step 12 told us to cut a Glama release by hand for every mcp release. Glama has no API and ignores the version typed into Create Release (it labelled 0.11.1 as 0.8.3), so we've stopped updating it. Step 12 now says so, and Auto-Release stays off.

Merge Danger

Door: two-way.

Blast Radius: docs.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg

Glama has no API and ignores the version typed into Create Release, so
its listing is no longer updated. Auto-Release stays off.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg
@vercel

vercel Bot commented Nov 3, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
wraps-web Ready Ready Preview Nov 3, 2026 4:14pm UTC
wraps-website Ready Ready Preview Nov 3, 2026 4:14pm UTC

Request Review

@stewartjarod
stewartjarod merged commit fa6b72b into main Nov 3, 2026
30 of 31 checks passed
@stewartjarod
stewartjarod deleted the advisor/release-drop-glama branch November 3, 2026 17:01
stewartjarod added a commit that referenced this pull request Nov 4, 2026
* fix(ci): gate Deploy API on the commit that was actually deployed

The gate diffed the tested commit against the head_sha of the last successful
Deploy API run. A run whose gate skipped still concludes success, and a
workflow_run run's head_sha is main's HEAD at trigger time, not the tested
commit. After the docs-only #248 skipped, #247 (API) compared bf89a46..bf89a46,
skipped too, and never deployed (runs 41681472352, 41683100866).

The deploy job now records the deployed commit as a success GitHub Deployment in
the api-<stage> environment after sst deploy, and the gate diffs against the
newest such record. The decision moves into scripts/ci/deploy-gate.sh with tests
that run it against a throwaway git repo and a stubbed gh. The first run after
this ships finds no record and deploys, which is the intended bootstrap.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg

* fix(ci): harden the Deploy API gate after review

- Never deploy backwards: skip a tested commit that is an ancestor of the
  recorded deploy (manual dispatch still always deploys).
- Add pnpm-lock.yaml and pnpm-workspace.yaml to the path list; the API takes its
  dependency versions from the pnpm catalog.
- Bound each gh call with timeout (DEPLOY_GATE_GH_TIMEOUT, default 20s) so a
  stalled API fails open; keep gh's stderr and say why each fail-open case
  deploys, with ::warning:: for API failures.
- git diff --no-renames so a file moved out of an API path still deploys.
- Fail with an explicit ::error:: when the gate script is missing from the
  tested commit, instead of a bare exit 127.
- Least-privilege job permissions: gate reads deployments, deploy writes them.
- Baseline tests pin the workflow half of the record: the tested sha, the
  api-<stage> environment after Deploy SST, POST only from deploy, and the
  removed runs lookup.
- Update docs/agents/ci-and-deploy.md for the script and the Deployment record.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg

* fix(ci): fail open when the gate cannot read a deployment status

If reading the newest record's status failed or timed out, the gate fell back to
the previous success record and diffed against a stale baseline. A revert of the
newest deploy then diffed as empty and was skipped, leaving the reverted code in
production. An unreadable status now deploys without a baseline; only a status
that was read and is not success is skipped in favour of the one before it.

- Warn on a deployments response without ids instead of the bootstrap notice,
  and include the gh exit code in the status-read warning.
- Annotate the downgrade guard's skip with how to roll back on purpose.
- Tests: a revert deploys, with the status read failing and with only the
  status call hanging; a revert of the recorded deploy deploys.
- Baseline: the deployment ref is built from DEPLOYED_SHA, the deploy job
  records under the gate's stage, and the record step has no if: or
  continue-on-error:. The comment on the record step says why a failed record
  keeps the job red.
- Reword ci-and-deploy.md: the no-backwards guarantee holds only when the
  record can be read; a manual dispatch is how to roll back on purpose.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg

* fix(ci): close the last two stale-baseline paths in the deploy gate

- A malformed deployment (no id or sha) is now a jq error, so the gate deploys
  without a baseline instead of silently dropping it and falling back to an older
  record. Drop the "without ids" branch it replaces.
- Split the status fetch from its parse: an empty (204) or non-array status body
  now deploys without a baseline instead of reading as "not success". The exit
  code in the warning is gh's own (1 or 124).
- Say that a rollback by dispatch lasts until the next push to main deploys, so
  revert on main to keep it (script comment and ci-and-deploy.md).
- Correct the record step's comment: a job can be re-run, a step cannot.
- Tests: empty status body, malformed newest deployment; pin the status-read
  warning by its exit code; document the stub's slow switches. Drop the
  duplicated skip line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GxJAQSsYCj3PjVtTjJLNWg

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

This branch was successfully deployed

2 active deployments
Preview – wraps-web — 7d1eab35 Deployed Nov 3, 2026 by vercel[bot]
Preview – wraps-website — 7d1eab35 Deployed Nov 3, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant