Improve release workflows?

[lolipopshock]

for example, when we approve a release workflow #403 , then we should automatically create tag and then we can use it to release a new version?

@sepo-agent /answer

[sepo-agent-app]

Yes, but I’d trigger it after the release PR is merged, not merely when the prepare workflow/PR is approved.

For #403 specifically, v0.3.1 already exists: https://github.com/self-evolving/repo/releases/tag/v0.3.1

Current release flow is intentionally “prepare only”:

• agent-release-prepare.yml creates/reuses a release issue, then dispatches implementation.
• agent-release.md tells the agent to update package/changelog but not create tags or releases.
• versioning.md documents that same boundary.
• release-version.ts already gives us the canonical version -> vX.Y.Z tag mapping.

Recommended plan:

1. Keep Agent / Release / Prepare as the agentic PR-creation step.
2. Add a deterministic Agent / Release / Publish workflow, source-repo gated like prepare.
3. Trigger publish on merged release PRs, ideally identified by a release marker in the PR body plus the changed files .agent/package.json and .agent/CHANGELOG.md.
4. In a typed CLI, read .agent/package.json, parse the version with parseReleaseVersion, verify the tag does not already exist, verify the changelog has notes for that version, then create the git tag and GitHub Release.
5. Add manual workflow_dispatch inputs for rerun/recovery, e.g. version, target_sha, and dry_run.
6. Update versioning.md so the documented flow becomes: prepare PR -> review/approve/merge -> publish workflow creates tag/release.

Main safety point: the tag should point at the merged default-branch commit, not the PR branch head. That makes the release reproducible and avoids publishing unmerged code.

If approved, I’d implement this as a small workflow plus CLI/tests, without changing the agent release prompt to perform GitHub writes directly.

---

codex | gpt-5.5/xhigh | Macmini-runner-1

[lolipopshock]

@sepo-agent /orchestrate

[sepo-agent-app]

Sepo is dispatching follow-up automation.

Source	Next	Target	Round	Status
orchestrate	implement	Issue #408	2 / 40	Dispatched

Reason: agent planner selected implement: Issue #408 has a bounded approved follow-up from the prior answer and no existing open PR; implement the release publish workflow directly on the issue.

[sepo-agent-app]

Sepo implementation finished

• Branch: agent/implement-issue-408/codex-26973791014
• Pull request: Add gated Sepo release publish workflow #411

Added a source-repo-gated Agent / Release / Publish workflow with a typed publish CLI that recognizes marked merged release PRs, validates required release files, package version, changelog notes, target SHA, and tag absence, then creates the GitHub Release/tag. Updated release PR prompting and versioning docs; verified with the agent build, focused release tests, docs validation, workflow YAML parsing, and git diff checks.

---

codex | gpt-5.5/xhigh | zucchini-m4-1

[github-actions]

Closing because this agent issue has had no activity for 30 days.