Command Reference

Command reference

Slash commands installed into .claude/commands/. Each is a markdown file: read it, change it, or delete it.

Command	Installed	What it does
/agent-permissions	always	Grant or revoke a single tool on a single roster seat, with the invariants that keep the roster safe enforced - reviewers stay read-only, only the orchestrator…
/board-audit	always	Find orphaned, stale, and loop-suspect work - Active tasks nobody is driving, finished subagent runs nobody logged, worktrees and branches the board does not k…
/brainstorm	needs long	Run a structured brainstorming session on a decision or a feature idea.
/code-graph	always	Build or refresh the code knowledge graph - modules, files, and import edges - that agents consult before any cross-module change.
/db-migration	needs db	Generate a database migration from schema changes, following the ERD and the data dictionary.
/deploy	always	Deploy to the hosting target.
/docs-graph	always	Build or refresh the DOCS knowledge graph - which documents define which IDs and who references whom - plus the interactive specs-graph.html.
/harness-toggle	always	Turn individual rules, commands, hooks, or agent seats off or back on at runtime - reversibly, with the change recorded and the harness graph refreshed.
/harness-tune	always	Adjust how tight the harness is after bootstrap - deployment rights, deny/ask lists, spawn allowlist, attempt caps - with every change shown as a diff and conf…
/harness-update	always	Re-run the harness bootstrap on this repo to pick up new assets, add or retire agents, or re-derive scopes after the codebase changed - reconciling, never clob…
/implement-fr	always	Plan and implement a functional requirement end-to-end against its acceptance criteria.
/new-adr	always	Create an Architecture Decision Record from the template.
/new-spec-section	always	Scaffold a specification section following the project's standard BA structure.
/new-task	always	Create a task file from the template and register it on the master plan, after the user confirms it should exist.
/review-changes	always	Run code review and security review on the current diff before opening a pull request.
/scaffold-feature	always	Create the skeleton of a feature module (entry point, library module, component).
/secret-scan	always	Scan the current changes for secrets and sensitive data before a commit or pull request.
/seed-db	needs db and db_seeder	Seed the local or development database with deterministic synthetic data per the ERD.
/skill-wire	always	Wire an installed skill to a roster seat after re-vetting its content - scope match, mandatory content review, invariant refusals, and a changelog record.
/sync-context	always	Update docs/context (glossary, business rules, known issues, tool changelog) from recent changes.
/task-resume	always	Resume a task from its task file after a compaction or in a new session.
/test	needs tests	Run lint plus the unit and end-to-end suites, and report each failure with its owning agent.

Turning things off

/harness-toggle moves a rule, command or hook into .claude/disabled/ and records the change in the committed .claude/disabled.json, including the hook's settings.json registration verbatim so enabling restores it byte-exactly. A re-run of the scaffolder respects it, so a toggled-off control is never silently resurrected.

Safety-critical controls need a typed confirmation phrase rather than a yes. Agents cannot toggle anything at all.