TRWDocumentation

Skills

Skills turn multi-step workflows into single commands. Type /trw-prd-new "Add rate limiting" and TRW handles the rest — create, groom, review, and plan in one pipeline. TRW ships 24 skills across five categories.

Sprint & Delivery

Manage the full sprint lifecycle from initialization through delivery. These skills coordinate PRD selection, task breakdown, agent assignment, and the delivery gate that ensures nothing ships without verification.

Command	What it does
/trw-sprint-init	Initialize a new sprint with PRD selection and tracking
/trw-sprint-finish	Complete sprint with build gate verification
/trw-sprint-team	Generate team playbooks for parallel agent work
/trw-deliver	Persist learnings and close the session
/trw-commit	Create a well-formatted git commit with type(scope) and PRD linking

Tip

/trw-sprint-team generates playbooks with exclusive file ownership for each agent. This prevents merge conflicts when running parallel agent teams.

Requirements

Author, groom, review, and plan PRDs. Each skill targets a specific stage of the requirements pipeline. The full pipeline runs from a rough feature idea to a sprint-ready execution plan with file ownership.

Command	What it does
/trw-prd-new	Create a new PRD and run the full readiness pipeline (groom, review, plan)
/trw-prd-ready	Full PRD lifecycle: create or pick up, groom, review, execution plan
/trw-prd-groom	Groom a PRD to sprint-ready quality
/trw-prd-review	Quality review with READY / NEEDS WORK verdict
/trw-exec-plan	Generate an execution plan from a groomed PRD

Tip

Start with /trw-prd-new for new features. It runs the entire pipeline end-to-end. Use individual skills like /trw-prd-groom only when you need to re-run a specific stage.

PRD Pipeline — one command, sprint-ready

Type /trw-prd-new "your feature" and TRW handles the rest. The full pipeline runs automatically:

terminal

CREATE → GROOM (≥0.85) → REVIEW (READY) → EXEC PLAN
                        ↑                    │
                        └── NEEDS WORK ──────┘
                            (max 2 loops)

example output

/trw-prd-new "Add webhook notifications for learning events"

→ Created PRD-CORE-082 (6 FRs, EARS format)
→ Grooming pass 1: score 0.72 → added traceability matrix
→ Grooming pass 2: score 0.88 ✓
→ Review: READY (no blocking issues)
→ Execution plan:
    Track A: src/webhooks/dispatcher.py (new)
    Track B: src/models/webhook_config.py (new)
    Track C: tests/test_webhooks.py (new)
    Shared:  src/config.py (modify — add webhook settings)

For existing PRDs, use /trw-prd-ready PRD-CORE-020 to pick up wherever the PRD is in the pipeline.

Memory & Knowledge

Inspect, curate, and optimize the learning store that compounds across sessions. Over time, learnings accumulate duplicates, stale entries, and tag imbalances. These skills keep the knowledge base healthy.

Command	What it does
/trw-memory-audit	Audit learning health: staleness, duplicates, tag distribution
/trw-memory-optimize	Prune stale entries, consolidate duplicates, rebalance tags
/trw-learn	Record a discovery for future sessions

Tip

Run /trw-memory-audit monthly. It surfaces stale entries that waste context tokens and duplicate learnings that should be consolidated.

Quality & Review

Run audits, reviews, and refactoring passes. These skills cover security, test strategy, code duplication, spec compliance, and team coordination. They are the second line of defense after automated build checks.

Command	What it does
/trw-audit	Adversarial spec-vs-code audit against PRD acceptance criteria
/trw-review-pr	Structured code review using FRAMEWORK.md rubric
/trw-simplify	Simplify code for clarity and maintainability
/trw-dry-check	Scan for duplicated code blocks
/trw-security-check	OWASP-focused security audit
/trw-test-strategy	Audit test coverage and suggest improvements
/trw-team-playbook	Generate teammate playbooks with file ownership

Warning

/trw-audit runs adversarial verification — it deliberately looks for gaps between your PRD and implementation. Expect it to find issues even in code that passes all tests. That is the point.

Framework & Health

Monitor ceremony compliance, phase gate status, and overall project health. Use these when onboarding to a project or when something feels off with the development workflow.

Command	What it does
/trw-framework-check	Check TRW ceremony compliance and phase gate status
/trw-project-health	Audit TRW health in the current project
/trw-ceremony-guide	Full ceremony and tool usage guide
/trw-email-template	Generate email templates

Info

/trw-project-health gives you a single-page summary of learning count, build pass rate, ceremony compliance, and configuration status. Start here when joining an existing TRW project.

Usage examples

Skills accept arguments inline. Here are common workflows showing what each skill actually produces.

Create a PRD from a feature idea

/trw-prd-new "Add webhook notifications for learning events"

→ Created PRD-CORE-082 with 6 EARS-format requirements
→ Grooming: 2 passes, final score 0.88/1.00
→ Review: READY (traceability 100%, density 92%)
→ Execution plan: 3 tracks, 4 files, ~2 hours estimated

Audit implementation against requirements

/trw-audit PRD-CORE-055

→ Mapped 8 FRs to implementation files
→ FR01-FR06: implemented and tested ✓
→ FR07: STUB — function exists but returns placeholder
→ FR08: MISSING — no implementation found
→ Verdict: 2 gaps found, blocking delivery

Start a sprint with team coordination

/trw-sprint-init

→ Selected PRDs: CORE-080, CORE-081, FIX-042
→ Task breakdown: 12 tasks across 3 tracks
→ Run directory: .trw/runs/sprint-55/
→ Ready for /trw-sprint-team to assign agents

Run a security audit

/trw-security-check

→ Scanning 78 modules for OWASP Top 10 patterns...
→ 0 critical, 1 high, 3 medium findings
→ HIGH: setattr from user input in admin.py:142
→ MED:  Missing rate limit on /auth/forgot-password
→ MED:  Broad exception handler hides errors in webhook.py
→ MED:  API key logged at DEBUG level in telemetry.py

Audit memory health

/trw-memory-audit

→ 163 active learnings (47 high-impact, 89 medium, 27 low)
→ 12 stale entries (no recall in 30+ days)
→ 8 near-duplicates detected
→ Tag distribution: testing (31), backend (28), security (19)...
→ Recommendation: run /trw-memory-optimize to prune 20 entries

Review changes before pushing

/trw-review-pr

→ Reviewing 4 changed files (187 lines added, 23 removed)
→ DRY: extracted shared helper — no duplicates ✓
→ KISS: complexity within bounds ✓
→ SOLID: single responsibility maintained ✓
→ Score: 8.2/10
→ 1 suggestion: consider adding error boundary in dispatcher.py:89

Creating custom skills

You can create project-specific skills that your AI agent discovers automatically. A skill is a directory inside .claude/skills/ containing a SKILL.md file.

directory structure

.claude/skills/
  my-deploy/
    SKILL.md          # Required — frontmatter + instructions
  my-db-migration/
    SKILL.md
    templates/        # Optional — supporting files

SKILL.md format

---
name: my-deploy
description: >
  Deploy the application to staging or production.
  Runs health checks after deployment.
user-invocable: true
argument-hint: "[staging|production]"
allowed-tools: Read, Bash, Grep
---

# My Deploy Skill

## Steps
1. Run pre-deploy checks (build, lint, test)
2. Deploy to the specified environment
3. Run health checks against the deployed URL
4. Report success or rollback on failure

Info

The frontmatter fields name, description, and user-invocable are required. Set user-invocable: true for skills the user types directly, or false for skills loaded automatically as context (like brand voice guidelines).

Skills vs Tools

Skills and tools serve different roles. Understanding the difference helps you know when to type a command versus letting your agent work automatically.

Aspect	Skills	Tools
Invocation	You type a slash command (`/trw-prd-new`)	Your AI calls them automatically
Who triggers	The developer	The AI agent
Scope	Multi-step workflows (orchestrate many tools)	Single operations (one function call)
Protocol	Markdown instructions in SKILL.md files	MCP (Model Context Protocol) functions
Customizable	Create your own in `.claude/skills/`	Built into the TRW MCP server
Example	`/trw-prd-new "feature"` runs create + groom + review + plan	`trw_prd_create` creates one PRD document

How to prompt

You don't need to memorize commands. Describe what you want in natural language and TRW suggests the right skill.

natural language mapping

"I'm done with this feature"           →  /trw-deliver
"Add user notifications"               →  /trw-prd-new
"Make PRD-CORE-020 sprint-ready"       →  /trw-prd-ready
"Review my changes before I push"      →  /trw-review-pr
"Start a new sprint"                   →  /trw-sprint-init
"Check for security issues"            →  /trw-security-check
"What has my AI learned so far?"       →  /trw-memory-audit
"Clean up duplicate learnings"         →  /trw-memory-optimize
"Is the PRD good enough to build?"     →  /trw-prd-review
"Split this into parallel agent work"  →  /trw-sprint-team
"Is everything working correctly?"     →  /trw-project-health

Next steps

Lifecycle

The six phases your AI moves through and when each skill fires

Tools

The MCP tools that skills orchestrate under the hood

Agents

Specialized roles for team-based sprints

Tools Agents