Skip to content

Skills Reference

Skills are slash commands invoked inside Claude Code. All 42 skills are embedded in the CQ binary (skills_embed build tag) — no internet required after install.

Ideation

SkillTriggersDescription
/piplay idea, ideation, brainstormBrainstorm and refine ideas before planning. Diverge/converge/research/debate modes. Writes idea.md and auto-launches /plan.

Core Workflow

SkillTriggersDescription
/planplan, planning, design, blueprintDiscovery → Design → Lighthouse contracts → Task creation. Full structured plan for a feature.
/runrun, execute, goSpawn workers for all pending tasks in parallel. Continuous mode — auto-respawns until queue empty.
/finishfinish, complete, wrap upBuild → test → docs → commit. Post-implementation completion routine.
/statusstatus, progressVisual task graph with progress, dependency graph, queue summary, and worker status.
/quickquick, fastCreate + assign one task immediately, skip planning. For small focused changes.

Quality Loop

SkillTriggersDescription
/polishpolish(Deprecated — polish loop is now built into /finish. No separate invocation needed.)
/refinerefine(Deprecated — quality loop is now built into /finish. No separate invocation needed.)
/checkpoint(auto at checkpoint)Phase gate: 4-lens review (holistic / user-flow / cascade / ship-ready). Approve, request changes, replan, or redesign.
/validatevalidate, checkRun lint + tests with severity-based handling. CRITICAL blocks commit, HIGH requires review, MEDIUM is recommended.
/reviewreviewComprehensive 3-pass code or paper review with 6-axis evaluation. Generates formal review document.
/company-reviewcompany review, code review, diff reviewPI Lab standard code review. PR/MR diff-based 6-axis evaluation.
/submitsubmit, sendSubmit completed task with automated validation. Verifies commit SHA, triggers checkpoint if needed.
/simplifysimplify, clean upReview changed code for reuse, quality, and efficiency, then fix any issues found.

Task Management

SkillTriggersDescription
/add-taskadd task, new taskAdd task interactively with DoD, scope, and domain guidance. Infers ID from existing patterns.
/stopstop, haltStop execution, transition to HALTED state. Preserves progress for later resumption.
/swarmswarmSpawn coordinator-led agent team. Modes: standard (implementation), review (read-only audit), investigate (hypothesis competition).

Session

SkillTriggersDescription
/donedone, session end, session doneMark the current session as done with full capture — summarize work, save knowledge, clean up state.
/attachsession name, attach, name this sessionAttach a name to the current session for later resume with cq claude -t <name>. Optionally add a memo.
/rebootreboot, restartReboot the current named session. cq resumes with the same session UUID automatically.
/session-distillsession distill, session summary, distillDistill the current session into persistent knowledge. Extracts decisions, patterns, and insights into the knowledge base.

Research & Documents

SkillTriggersDescription
/c9-initc9-init, c9 initializeInitialize a new C9 research project. Creates state.yaml with metric, convergence conditions, and Hub URL.
/c9-loopc9-loopMain loop driver — reads current phase from state.yaml and auto-executes next step.
/c9-surveyc9-surveySurvey latest arXiv papers + SOTA benchmarks using Gemini Google Search grounding.
/c9-conferencec9-conferenceClaude (Opus) + Gemini (Pro) debate mode — research conference simulation.
/c9-steerc9-steerChange phase and update reason without editing state.yaml directly.
/c9-reportc9-reportCollect experiment results from remote server via distributed worker.
/c9-finishc9-finishSave best model + document results when research loop completes.
/c9-deployc9-deployDeploy best model to edge server. Can run independently of /c9-finish.
/research-loopresearch loopPaper-experiment improvement loop. Iterates review → plan → experiment → re-review until target quality reached.
/experiment-workflowexperiment workflow, experiment lifecycleEnd-to-end experiment lifecycle management: data prep → train → eval → record.
/c2-paper-reviewpaper review, academic review(Deprecated — use /review instead.)

Development

SkillTriggersDescription
/tdd-cycleTDD, test-driven, RED-GREEN-REFACTORTDD cycle guide. RED-GREEN-REFACTOR sequence for test-driven development.
/debuggingdebug, debugging, bug trackingSystematic debugging. Reproduce → hypothesis → isolate → fix → verify sequence.
/spec-firstspec-first, spec first, design docSpec-First development guide. Write spec doc before code.
/incident-responseincident, outage, server down, error rate surgeProduction incident response workflow. Triage → diagnose → mitigate → postmortem.

Meta & Utilities

SkillTriggersDescription
/craftcraft, create skill, create ruleInteractively create skills, agents, rules, and CLAUDE.md customizations.
/helphelpQuick reference for skills, agents, and MCP tools. Decision tree + keyword search across all 42 skills.
/clearclearReset C4 state for debugging. Clears tasks, events, locks in .c4/ with optional config preservation.
/initinit, initializeInitialize C4 in current project. Detects installation path, runs cq claude/cursor/codex.
/claude-md-improverimprove CLAUDE.md, claude-md, improve instructionsAnalyze and improve the project's CLAUDE.md. Structure check, build/test commands, agent rules.
/skill-testerskill tester, skill eval, evalTest and evaluate skill quality. Generate eval cases, run classification trials, score trigger accuracy.
/pr-reviewcreate PR, PR checklist, pull requestPR/MR creation checklist and review guide. Auto-validates before merge.
/releasereleaseGenerate CHANGELOG from git history. Conventional Commits analysis, semantic version suggestion, tag creation.
/standbystandby, wait, worker modeConvert session into persistent distributed worker via Supabase. Waits for jobs, executes, reports back. full tier only
/interviewinterviewDeep exploratory requirements interview. Acts as senior PM/architect to discover hidden requirements and edge cases.

Skill Health

Requires connected or full tier (LLM Gateway needed for haiku classification).

Measure and monitor whether skills trigger correctly — ensuring Claude classifies user prompts accurately before and after changes.

MCP ToolDescription
c4_skill_eval_runRun k-trial haiku classification on a skill's EVAL.md test cases. Returns trigger_accuracy.
c4_skill_eval_generateGenerate EVAL.md test cases (positive + negative prompts) for a skill using haiku.
c4_skill_eval_statusShow trigger accuracy summary for all evaluated skills. ok = ≥ 0.90.

cq doctor includes a skill-health check that warns when any skill drops below the 0.90 threshold.


Machine-readable

Download as JSONL for programmatic use:

sh
curl https://playidea-lab.github.io/cq/api/skills.jsonl