- Updated command syntax formatting to use code blocks for clarity in `prep.md`, `review.md`, and `spec.md`. - Enhanced architectural diagrams in `ch01-what-is-claude-dms3.md` and core concepts in `ch03-core-concepts.md` using mermaid syntax for better visualization. - Improved workflow diagrams in `ch04-workflow-basics.md` and `4-level.md` to provide clearer representations of processes. - Added troubleshooting section in `installation.md` to address common installation issues and provide quick start examples. - Revised skill documentation in `claude-meta.md` and `claude-workflow.md` to standardize command triggers and output structures. - Updated best practices and workflow index documentation to enhance readability and understanding of workflow levels and practices.
14 KiB
Codex Skills Overview
One-Liner
Codex Skills is a specialized skill system for the Codex model — implementing multi-agent parallel development and collaborative analysis through lifecycle, workflow, and specialized skill categories.
vs Claude Skills Comparison
| Dimension | Claude Skills | Codex Skills |
|---|---|---|
| Model | Claude model | Codex model |
| Architecture | team-worker agent architecture | spawn-wait-close agent pattern |
| Subagents | discuss/explore subagents (inline calls) | discuss/explore subagents (independent calls) |
| Coordinator | Built-in coordinator + dynamic roles | Main process inline orchestration |
| State Management | team-session.json | state file |
| Cache | explorations/cache-index.json | shared discoveries.ndjson |
Skill Categories
| Category | Document | Description |
|---|---|---|
| Lifecycle | lifecycle.md | Full lifecycle orchestration |
| Workflow | workflow.md | Parallel development and collaborative workflows |
| Specialized | specialized.md | Specialized skills |
Core Concepts Overview
| Concept | Description | Location/Command |
|---|---|---|
| team-lifecycle | Full lifecycle orchestrator | /team-lifecycle |
| parallel-dev-cycle | Parallel development cycle | /parallel-dev-cycle |
| analyze-with-file | Collaborative analysis | /analyze-with-file |
| brainstorm-with-file | Brainstorming | /brainstorm-with-file |
| debug-with-file | Hypothesis-driven debugging | /debug-with-file |
Lifecycle Skills
team-lifecycle
One-Liner: Full lifecycle orchestrator — spawn-wait-close pipeline for spec/implementation/test
Triggers:
/team-lifecycle <task-description>
Features:
- 5-phase pipeline: requirements clarification → session initialization → task chain creation → pipeline coordination → completion report
- Inline discuss: Production roles (analyst, writer, reviewer) inline invoke discuss subagents, reducing spec pipeline from 12 beats to 6
- Shared explore cache: All agents share centralized
explorations/directory, eliminating duplicate codebase exploration - Fast-advance spawning: Immediately spawn next linear successor after agent completion
- Consensus severity routing: Discussion results routed through HIGH/MEDIUM/LOW severity levels
Agent Registry:
| Agent | Role | Mode |
|---|---|---|
| analyst | Seed analysis, context collection, DISCUSS-001 | 2.8 Inline Subagent |
| writer | Documentation generation, DISCUSS-002 to DISCUSS-005 | 2.8 Inline Subagent |
| planner | Multi-angle exploration, plan generation | 2.9 Cached Exploration |
| executor | Code implementation | 2.1 Standard |
| tester | Test-fix loop | 2.3 Deep Interaction |
| reviewer | Code review + spec quality, DISCUSS-006 | 2.8 Inline Subagent |
| architect | Architecture consultation (on-demand) | 2.1 Standard |
| fe-developer | Frontend implementation | 2.1 Standard |
| fe-qa | Frontend QA, GC loop | 2.3 Deep Interaction |
Pipeline Definition:
Spec-only (6 beats):
RESEARCH-001(+D1) → DRAFT-001(+D2) → DRAFT-002(+D3) → DRAFT-003(+D4) → DRAFT-004(+D5) → QUALITY-001(+D6)
Impl-only (3 beats):
PLAN-001 → IMPL-001 → TEST-001 || REVIEW-001
Full-lifecycle (9 beats):
[Spec pipeline] → PLAN-001 → IMPL-001 → TEST-001 || REVIEW-001
Beat Cycle:
event (phase advance / user resume)
↓
[Orchestrator]
+-- read state file
+-- find ready tasks
+-- spawn agent(s)
+-- wait(agent_ids, timeout)
+-- process results
+-- update state file
+-- close completed agents
+-- fast-advance: spawn next
+-- yield (wait for next event)
Session Directory:
.workflow/.team/TLS-<slug>-<date>/
├── team-session.json # Pipeline state
├── spec/ # Specification artifacts
├── discussions/ # Discussion records
├── explorations/ # Shared exploration cache
├── architecture/ # Architecture assessments
├── analysis/ # Analyst design intelligence
├── qa/ # QA audit reports
└── wisdom/ # Cross-task knowledge accumulation
parallel-dev-cycle
One-Liner: Multi-agent parallel development cycle — requirements analysis, exploration planning, code development, validation
Triggers:
/parallel-dev-cycle TASK="Implement feature"
/parallel-dev-cycle --cycle-id=cycle-v1-20260122-abc123
/parallel-dev-cycle --auto TASK="Add OAuth"
Features:
- 4 specialized workers: RA (requirements), EP (exploration), CD (development), VAS (validation)
- Main process inline orchestration (no separate orchestrator agent)
- Each agent maintains one main document (complete rewrite per iteration) + auxiliary log (append)
- Automatically archive old versions to
history/directory
Workers:
| Worker | Main Document | Auxiliary Log |
|---|---|---|
| RA | requirements.md | changes.log |
| EP | exploration.md, architecture.md, plan.json | changes.log |
| CD | implementation.md, issues.md | changes.log, debug-log.ndjson |
| VAS | summary.md, test-results.json | changes.log |
Shared Discovery Board:
- All agents share real-time discovery board
coordination/discoveries.ndjson - Agents read at start, append discoveries during work
- Eliminates redundant codebase exploration
Session Structure:
{projectRoot}/.workflow/.cycle/
├── {cycleId}.json # Main state file
├── {cycleId}.progress/
│ ├── ra/ # RA agent artifacts
│ │ ├── requirements.md # Current version (complete rewrite)
│ │ ├── changes.log # NDJSON full history (append)
│ │ └── history/ # Archived snapshots
│ ├── ep/ # EP agent artifacts
│ ├── cd/ # CD agent artifacts
│ ├── vas/ # VAS agent artifacts
│ └── coordination/ # Coordination data
│ ├── discoveries.ndjson # Shared discovery board
│ ├── timeline.md # Execution timeline
│ └── decisions.log # Decision log
Execution Flow:
Phase 1: Session initialization
↓ cycleId, state, progressDir
Phase 2: Agent execution (parallel)
├─ Spawn RA → EP → CD → VAS
└─ Wait for all agents to complete
Phase 3: Result aggregation & iteration
├─ Parse PHASE_RESULT
├─ Detect issues (test failures, blockers)
├─ Issues AND iteration < max?
│ ├─ Yes → Send feedback, loop back to Phase 2
│ └─ No → Proceed to Phase 4
└─ Output: parsedResults, iteration status
Phase 4: Completion and summary
├─ Generate unified summary report
├─ Update final state
├─ Close all agents
└─ Output: Final cycle report
Version Control:
- 1.0.0: Initial cycle → 1.x.0: Each iteration (minor version increment)
- Each iteration: Archive old version → Complete rewrite → Append changes.log
Workflow Skills
analyze-with-file
One-Liner: Collaborative analysis — interactive analysis with documented discussions, inline exploration, and evolving understanding
Triggers:
/analyze-with-file TOPIC="<question>"
/analyze-with-file TOPIC="--depth=deep"
Core Workflow:
Topic → Explore → Discuss → Document → Refine → Conclude → (Optional) Quick Execute
Key Features:
- Documented discussion timeline: Capture understanding evolution across all phases
- Decision logging at every key point: Force recording of key findings, direction changes, trade-offs
- Multi-perspective analysis: Support up to 4 analysis perspectives (serial, inline)
- Interactive discussion: Multi-round Q&A, user feedback and direction adjustment
- Quick execute: Direct conversion of conclusions to executable tasks
Decision Recording Protocol:
| Trigger | Content to Record | Target Section |
|---|---|---|
| Direction choice | Choice, reason, alternatives | #### Decision Log |
| Key findings | Finding, impact scope, confidence | #### Key Findings |
| Assumption change | Old assumption → New understanding, reason, impact | #### Corrected Assumptions |
| User feedback | User's raw input, adoption/adjustment reason | #### User Input |
brainstorm-with-file
One-Liner: Multi-perspective brainstorming — 4 perspectives (Product, Technical, Risk, User) parallel analysis
Triggers:
/brainstorm-with-file TOPIC="<idea>"
Features:
- 4-perspective parallel analysis: Product, Technical, Risk, User
- Consistency scoring and convergence determination
- Feasibility recommendations and action items
Perspectives:
| Perspective | Focus Areas |
|---|---|
| Product | Market fit, user value, business viability |
| Technical | Feasibility, tech debt, performance, security |
| Risk | Risk identification, dependencies, failure modes |
| User | Usability, user experience, adoption barriers |
debug-with-file
One-Liner: Hypothesis-driven debugging — documented exploration, understanding evolution, analysis-assisted correction
Triggers:
/debug-with-file BUG="<bug description>"
Core Workflow:
Explore → Document → Log → Analyze → Correct Understanding → Fix → Verify
Key Enhancements:
- understanding.md: Timeline of exploration and learning
- Analysis-assisted correction: Verify and correct assumptions
- Consolidation: Simplify proven-misunderstood concepts to avoid confusion
- Learning preservation: Retain insights from failed attempts
Session Folder Structure:
{projectRoot}/.workflow/.debug/DBG-{slug}-{date}/
├── debug.log # NDJSON log (execution evidence)
├── understanding.md # Exploration timeline + consolidated understanding
└── hypotheses.json # Hypothesis history (with determination)
Modes:
| Mode | Trigger Condition |
|---|---|
| Explore | No session or no understanding.md |
| Continue | Session exists but no debug.log content |
| Analyze | debug.log has content |
collaborative-plan-with-file
One-Liner: Collaborative planning — multi-agent collaborative planning (alternative to team-planex)
Triggers:
/collaborative-plan-with-file <task>
Features:
- Multi-agent collaborative planning
- planner and executor work in parallel
- Intermediate artifact files pass solution
unified-execute-with-file
One-Liner: Universal execution engine — alternative to workflow-execute
Triggers:
/unified-execute-with-file <session>
Features:
- Universal execution engine
- Support multiple task types
- Automatic session recovery
roadmap-with-file
One-Liner: Requirement roadmap planning
Triggers:
/roadmap-with-file <requirements>
Features:
- Requirement to roadmap planning
- Priority sorting
- Milestone definition
review-cycle
One-Liner: Review cycle (Codex version)
Triggers:
/review-cycle <target>
Features:
- Code review
- Fix loop
- Verify fix effectiveness
workflow-test-fix-cycle
One-Liner: Test-fix workflow
Triggers:
/workflow-test-fix-cycle <failing-tests>
Features:
- Diagnose test failure causes
- Fix code or tests
- Verify fixes
- Loop until passing
Specialized Skills
clean
One-Liner: Intelligent code cleanup
Triggers:
/clean <target>
Features:
- Automated code cleanup
- Code formatting
- Dead code removal
csv-wave-pipeline
One-Liner: CSV wave processing pipeline
Triggers:
/csv-wave-pipeline <csv-file>
Features:
- CSV data processing
- Wave processing
- Data transformation and export
memory-compact
One-Liner: Memory compression (Codex version)
Triggers:
/memory-compact
Features:
- Memory compression and merging
- Clean redundant data
- Optimize storage
ccw-cli-tools
One-Liner: CLI tool execution specification
Triggers:
/ccw-cli-tools <command>
Features:
- Standardized CLI tool execution
- Parameter specification
- Unified output format
issue-discover
One-Liner: Issue discovery
Triggers:
/issue-discover <context>
Features:
- Discover issues from context
- Issue classification
- Priority assessment
Related Documentation
Best Practices
-
Choose the right team type:
- Full lifecycle →
team-lifecycle - Parallel development →
parallel-dev-cycle - Collaborative analysis →
analyze-with-file
- Full lifecycle →
-
Leverage Inline Discuss: Production roles inline invoke discuss subagents, reducing orchestration overhead
-
Shared Cache: Utilize shared exploration cache to avoid duplicate codebase exploration
-
Fast-Advance: Linear successor tasks automatically skip orchestrator, improving efficiency
-
Consensus Routing: Understand consensus routing behavior for different severity levels
Usage Examples
# Full lifecycle development
/team-lifecycle "Build user authentication API"
# Parallel development
/parallel-dev-cycle TASK="Implement notifications"
# Collaborative analysis
/analyze-with-file TOPIC="How to optimize database queries?"
# Brainstorming
/brainstorm-with-file TOPIC="Design payment system"
# Debugging
/debug-with-file BUG="System crashes intermittently"
# Test-fix
/workflow-test-fix-cycle "Unit tests failing"
Statistics
| Category | Count |
|---|---|
| Lifecycle | 2 |
| Workflow | 8 |
| Specialized | 6 |
| Total | 16+ |