Produce a prioritized, read-only audit of the workspace's health and coherence (code, docs, integrity, standards) as a dated report. Use periodically or after big changes to surface what to improve. Read-only — it reports and recommends, never fixes.
Configure or re-configure this AI workspace: analyze an existing repo (or set up a new one) and propose a workspace.config.yaml + skill set. Trigger: when the user wants to set up, configure, or re-detect the workspace.
Assess feasibility and security of version bumps/migrations before touching anything. Trigger: when asked to update dependencies, migrate, or resolve version conflicts.
Keep the living docs (project status) current so the AI always has fresh project context. Trigger: after finishing a task or when project state changed.
Implement the tasks in order, keeping the code aligned with the spec and design, checking each off as it lands. Use to execute an agreed plan — if reality diverges from the spec, stop and update the spec first.
Fold a verified change's delta into the stable specs and move its folder to the archive. Use only after verify passes: ADDED requirements are appended, MODIFIED replace, REMOVED delete; then the change is archived with its full history.
Surface and resolve ambiguity before the spec is written: ask targeted questions about underspecified behaviour, edge cases and decisions, and record the answers. Use between proposal and spec whenever the proposal leaves choices open.
Decide HOW the spec is met: the technical approach, the key architecture decisions and their trade-offs, with Mermaid diagrams where they clarify. Use after the spec is stable, before breaking work into tasks.