Release & Deploy (GitHub Tags)

Leading words: commits ≠ deploys; branch → channel; tag triggers ship; changelog is truth; interview once; never guess channel.

Core model

Commits ≠ deploys. Only a git tag + GitHub Release triggers deployment (via GitHub Actions or project-specific release jobs).
Branch → channel
- main → production release
- staging → beta release
Tags
- Production: vMAJOR.MINOR.PATCH (e.g. v1.4.0)
- Beta: vMAJOR.MINOR.PATCH-beta.N (e.g. v1.4.0-beta.1)
Release notes come from the project's maintained CHANGELOG.md (do not invent a changelog). See references/versioning.md.

Load project config
- Look for projects/<repo-slug>.md next to this SKILL.md (slug = GitHub owner/repo with / → -, e.g. acme-web-app).
- If missing → run the first-time interview in references/interview.md, write the file from references/project.template.md, then continue.
Confirm channel with the user: beta (staging) or production (main). Never guess.
Verify git state (PowerShell-safe; run as separate commands):
- Correct branch checked out (staging or main)
- Working tree clean (or user explicitly accepts dirty)
- Branch is up to date with its remote
- HEAD is the commit they intend to ship
Determine next version from existing tags + user intent. Follow references/versioning.md.
Extract notes from CHANGELOG.md for that version (or [Unreleased] if they are cutting that section into a version). If changelog is missing/ambiguous → stop and ask.

Ensure on the correct branch at the intended commit.
Create annotated tag locally (optional if gh release create will create it).
gh release create with tag, title, notes from changelog, and --target set to the branch (or SHA).
Watch the project's release/deploy workflow (from project config) until green or report failure.
Summarize: tag, release URL, channel, deploy target/URL from project config.

Never tag production from staging, or beta from main, unless the project config explicitly documents a different branch map and the user confirms.
Never force-push tags or delete remote tags without explicit user request.
Never skip the interview on a first release in a repo.
Never fabricate changelog entries; only use CHANGELOG.md (or ask the user to update it first).
Prefer gh over raw API calls. Prefer project config over inventing deploy steps.

After the interview, project config names the stack. Read the matching reference only when needed:

Cloudflare Workers/Pages → references/platforms/cloudflare.md
Desktop apps (installers, auto-update, signed builds) → references/platforms/desktop.md
Generic / other → rely on projects/<slug>.md workflow names and URLs