What problem does it solve? Deciding whether, what, and when to release is error-prone: agents and humans conflate a release recommendation with the authority to cut one, leading to unapproved version bumps, commits, and tags. This skill separates the release decision from publication by producing an evidence-backed recommendation and enforcing explicit human approval gates before any release artifact is created. ## Core Features & Use Cases - Evidence-based release recommendation: Gathers merged PRs and linked issues since the latest tag via bin/release-evidence.py, classifies changes as none/patch/minor/breaking/uncertain, and recommends a version and timing with rationale and confidence. - Hard publication boundary: Never bumps VERSION, edits CHANGELOG.md, commits, tags, publishes, or deploys; two explicit approval gates (with an operator-supplied approval token) guard the dry-run and apply steps of the configured release strategy. - Inherited policy deference: Detects a .domi-pin via bin/domi-status.sh and routes the version/timing call upstream to DomI policy instead of settling it locally, with script-enforced stop conditions (exit codes 64/65/66). - Use Case: A maintainer asks "should we release this, and take care of it?" The skill gathers evidence since the last tag, recommends v1.3.0 with release-now timing and confidence, then stops — creating the Release Please PR only after explicit human approval at each gate. ## Quick Start Ask the agent "should we cut a release, and if so what version?" and it will gather evidence since the latest tag and return a version and timing recommendation without tagging or publishing anything.
