Decomposes a spec or architecture into buildable tasks with acceptance criteria, dependencies, and risk-first implementation order, for AI agents or human engineers. Produces `docs/forsvn/artifacts/meta/tasks.md`. Use once requirements are clear and you need an execution plan. Not for clarifying unclear requirements (use discover) or designing architecture (use architect-system). For code review after building, see review-work.
Designs technical blueprints — tech stack selection, database schema, API design, file structure, and deployment plan for a defined product or feature. Produces `architecture/system-architecture.md`. Not for unclear requirements (use discover) or task decomposition (use breakdown-tasks). For user journey mapping, see map-user-flow. For code quality after building, see review-work.
Designs the in-product interface from a validated map-user-flow — screen inventory, a reusable component system, DESIGN-token application, per-state visual treatments, accessibility floor, and a buildable handoff. Emits a portable SPEC, never rendered UI. Use after a user flow exists and before architect-system. Not for brand identity (create-brand), conversion landing pages (brief-landing-page), or technical architecture (architect-system).
Audits and refactors existing code for readability, maintainability, and dead code removal without changing behavior. Produces `docs/forsvn/artifacts/meta/records/[date]-cleanup-<slug>.md` and applies fixes in-place. Not for diagnosing business problems (use diagnose) or writing documentation (use write-docs). For writing missing docs after cleanup, see write-docs.
Audit + clean a developer's machine — dotfolders, caches, language toolchains, package globals — with per-target classification, risk surfacing (auth, processes, side effects), and explicit confirmation. Produces `docs/forsvn/artifacts/meta/records/machine-cleanup-*.md`. Not for cleaning code (use clean-code) or for triaging user files (those need human review).
Extracts repeated operational mechanics (SDK, API, file-system, or network logic copy-pasted across handlers or actions) into a shared service layer — produces a stepwise migration plan, then applies it caller-by-caller with verification at each step. Use when the same "how" recurs across 2+ callers. Not for behavioral cleanup or dead code (use clean-code) or designing a system from scratch (use architect-system). For a quality review after migrating, see review-work.
Maps multi-step in-product flows — screens, decisions, transitions, platform-native touchpoints (dock, menu bar, widgets, notifications, Live Activity), edge cases, and error states for a feature or user journey. Use when designing or auditing a feature spanning multiple screens, states, or platforms. Not for visual brand design (use create-brand), landing-page architecture (use brief-landing-page), or technical architecture (use architect-system).
Generate documentation from a codebase — READMEs, API references, setup guides, runbooks, architecture docs, ship logs, release notes (CHANGELOG + GitHub Release bodies). Not for specifying what to build (use discover), restructuring code (use clean-code), or task decomposition (use breakdown-tasks).