Release workflow (/release) is Claude Code-only, inaccessible to other agents/humans

[markrichardson]

Summary

The documented release workflow (/release) is implemented as a Claude Code slash-command
(rhiza-claude), with no equivalent for other coding agents / assistants (e.g. GitHub Copilot,
Cursor, plain CLI users). This makes the release process effectively inaccessible to anyone not
using Claude Code, even though the underlying steps (version bump, changelog generation, branch/PR,
tag push) are not inherently Claude-specific.

Where this shows up

• .github/workflows/rhiza_release.yml (synced into downstream repos) contains this comment:

  # 📄 CHANGELOG: not updated here. The release process (rhiza-claude `/release`)
  # folds a freshly generated CHANGELOG.md into the version-bump commit before the
  # tag is pushed, so the tagged commit already carries the changelog — no separate
  # post-tag commit.
  

  This makes the workflow's correctness depend on a manual step that only exists as a
  Claude Code command, with no documented tool-agnostic equivalent (e.g. a plain script or
  make target that any agent or human could run).

• Downstream, we ended up writing our own from-scratch release runbook (an agent "skill" file)
  to reconstruct the steps Claude's /release presumably performs, purely so other
  assistants/agents (and humans) have something to follow.

Request

Could the release process be made agent-agnostic? For example:

• Extract the /release logic (version bump, changelog generation via git-cliff or similar,
  branch/PR creation, tag push) into a plain script or make/uvx target that lives in the repo,
  rather than only as a Claude Code slash-command.
• Have rhiza-claude's /release simply invoke that script, so Claude Code, other agents, and
  humans all follow the same documented path.
• Document the expected pre-tag state (e.g. "CHANGELOG.md must already reflect this release before
  tagging") independent of any particular assistant.

Happy to share our SKILL.md reconstruction as a starting point if useful.

[markrichardson]

cf. #1654