Implement OpenSpec Changes

Run apply → verify → fix → goodreview → fix → commit for each change, sequentially.

Setup

Usage: /implement-openspec-changes [change-id ...]

• Without IDs, list active changes (openspec list --json, or inspect openspec/changes/ excluding archive/) and ask which to implement. Validate all supplied IDs before starting; stop on missing changes.
• Locate openspec-apply-change, openspec-verify-change, goodreview, and git in the runtime's installed skills, project/user skill directories, or this repository's skills/. Stop and name missing dependencies. OpenSpec supplies the first two; GoodSkills supplies the others.
• Before each change, record HEAD, staged/unstaged diffs, and untracked files. Track this change's files and hunks against that baseline; preserve unrelated work and its staging state. If ownership overlaps ambiguously, ask before modifying or committing those hunks.

Execution

The host orchestrates. Announce <change-id> · N/6 · <step> before launching one generic subagent per step in the shared workspace. Wait for completion before advancing; no isolated worktrees or named custom agents. If subagents are unavailable, execute locally with the same gates and label the fallback.

Give each child the repository path, change ID, scope/baseline, these execution/repair gates, relevant dependency skill paths, and preceding reports needed for its step. Require it to read the skill and referenced resources. Only apply/fix children may edit; only the commit child may stage or commit. Each returns status, findings/dispositions, changed files, checks/results, and blockers as applicable. Missing or failed reports are blockers, not clean results.