Working tree clean — git status --porcelain returns empty. Uncommitted work first, then release.
Tests pass — bats tests/ succeeds. Stop on failure; failing tests are never a "release later" problem.
ShellCheck clean — shellcheck -x -S warning -e SC1091 bin/agentsync.sh lib/**/*.sh returns 0.
CHANGELOG covers all user-facing commits since the last tag:
- git log --oneline $(git describe --tags --abbrev=0)..HEAD lists the candidates.
- For each commit, decide: user-visible (CLI flag, output format, new tool, bug a user could hit) → must appear in the new CHANGELOG section. Internal refactor / test-only / CI / dev tooling → stays out.
- If a user-visible commit is missing from your draft, add it before bumping.
Docs reflect current behaviour — spot-check that the release does not ship with stale docs:
- README.md — command list, flag list, supported tools table. If a command/flag was renamed, removed, or added since the last tag, README must match.
- .ai/src/commands/*.md and .ai/src/skills/**/SKILL.md — descriptions, argument hints, examples line up with what bin/agentsync.sh actually accepts.
- .ai/src/rules/*.md — invariants and module map (architecture.md) match current file layout in lib/helpers/.
- .ai/src/tools/_TEMPLATE.yaml — every YAML option referenced by lib/sync.sh is documented; no documented option is dead.
- Cross-check git log --since="<last-tag-date>" --name-only against the doc files: if lib/sync.sh or bin/agentsync.sh changed and no doc file did, ask whether docs need a follow-up before tagging.

When a doc gap is real, fix it in the same release commit (or a separate commit immediately before) rather than punting to a "docs" release later — release notes that lie age the project faster than missing features.

Steps

Pre-release checks — Run the five checks above. Stop on the first failure and report.
Determine bump type — major (breaking changes), minor (new features), patch (bug fixes). Default: patch.
Calculate new version — Parse MAJOR.MINOR.PATCH from VERSION, increment the appropriate component.
Write CHANGELOG entry — Add ## X.Y.Z section at the top of CHANGELOG.md (after # Changelog):
- Group under ### Added, ### Changed, ### Fixed, ### Removed.
- Bold the feature/component name: - **Export command:** added --dry-run support.
- Describe user-facing impact, not implementation details.
- Match the tone and format of existing entries.
Update VERSION — Write the new version number to VERSION (no v prefix, no trailing newline).
Commit — git add VERSION CHANGELOG.md <any doc files fixed above> && git commit -m "release: vX.Y.Z".
Leave the push to the user — CI handles the rest:
- auto-tag.yaml creates the git tag when VERSION changes on main.
- release.yaml creates the GitHub Release using the CHANGELOG section as release notes.

CHANGELOG Format

## 0.5.0

### Added
- **Feature name:** Description of what was added and why it matters.

### Changed
- **Component:** What changed and how it affects users.

### Fixed
- **Bug description:** What was broken and how it's fixed now.

Gotchas

Write the VERSION file as 0.4.3, never v0.4.3 — the v prefix breaks the auto-tag workflow.
Leave git tag creation to CI. auto-tag.yaml runs when VERSION changes on main.
Stop after the commit — let the user review before pushing.
CHANGELOG entries extracted by CI use awk with exact ## X.Y.Z matching — the version header must be ## X.Y.Z with no extra text.
The agentsync release CLI command exists for maintainer use with interactive confirmation. The AI workflow edits files and commits.
Keep CHANGELOG entries to user-visible changes. Internal refactors with no behavioural effect stay out.
Doc fixes uncovered by the pre-release checks ride with the release commit; never ship a known-stale README "to fix next time".

release

Release

Pre-release checks