Use when a maintainer needs to cut, bump, or verify a repo release surface such as plugin versions, generated install manifests, and operator update instructions.
Use when a non-trivial design decision, code change, release, rename, deletion, spec, or workflow change needs a before-the-fact critique. Probe distinct failure angles, then run a counterweight pass that separates real blockers from over-worry before the change locks in.
Use after a meaningful work unit or when the user asks for a retrospective. Reviews what happened, what created waste, which decisions mattered, which named expert lens or direct counterfactual would have changed the next move, and which workflow/capability/memory improvements should make the next session better. Auto-selects `session` or `weekly` mode from context; ambiguous cases default to `session`.
Use when the goal is to understand and improve the repo's current quality bar. Detect existing gates, run the available ones, inspect concept integrity, test confidence, and security posture, then propose concrete quality moves instead of only complaining about what is missing.
Use when filing a GitHub issue from current context or resolving GitHub issues end-to-end through the adapter-resolved backend (`gh` by default, or a host-mediated capability such as `acme github`). Issue creation reports the observed problem before suggesting solutions; issue resolution treats GitHub as the source of truth, classifies the issue, runs a causal review for bug-class issues before designing the fix, and runs a resolution critique so the same class of issue does not recur.
Use when a repo needs its initial operating surface created or normalized. Bootstrap README, AGENTS.md, CLAUDE.md symlink policy, roadmap, and operator-acceptance docs from minimal ideation for greenfield repos, or realign those same surfaces for partially-initialized repos without pretending quality review or deep product ideation already happened.
Use when a concept needs to become a living implementation contract. Refine ideation artifacts or existing design docs into the current build contract, decide what must be fixed now versus probed during implementation, define testable success criteria, and keep the contract synchronized as `impl` learns new facts.
Use when creating or upgrading a repo-owned CLI, bootstrap script, or command runner. Define the command surface, install/update contract, structured output, dry-run and doctor behavior, distribution path, and quality gates before spreading ad hoc shell or Python entrypoints.