GOAP Agent Skill: Goal-Oriented Action Planning

Enable intelligent planning and execution of complex multi-step tasks through systematic decomposition, dependency mapping, and coordinated multi-agent execution.

Always use the plans/ folder for all files.

Quick Reference

See the GOAP Agent documentation (../agents/goap-agent.md) for detailed execution patterns and examples.

CRITICAL: Understanding Skills vs Task Agents

There are TWO different types of workers you can coordinate in GOAP planning:

Skills (invoked via Skill tool)

Skills are instruction sets that guide Claude directly. They provide specialized knowledge and workflows.

How to invoke: Skill(command="skill-name")

When to use:

• •Need specialized knowledge/workflow guidance
• •Task requires deep domain expertise (Rust quality, architecture validation)
• •Want to follow a proven methodology
• •Examples: Code quality review, gap analysis, architecture validation

Task Agents (invoked via Task tool)

Task Agents are autonomous sub-processes that execute tasks independently using tools.

How to invoke: Task(subagent_type="agent-name", prompt="...", description="...")

When to use:

• •Need autonomous task execution
• •Task requires tool usage (Read, Edit, Bash, etc.)
• •Want parallel/independent execution
• •Examples: Running tests, implementing features, debugging

Common Error to Avoid

WRONG: Task(subagent_type="rust-code-quality", ...) → ERROR! rust-code-quality is a Skill!

CORRECT: Skill(command="rust-code-quality") → SUCCESS

See the agent-coordination skill for the complete reference on Skills vs Agents.

When to Use This Skill

Use this skill when facing:

• •Complex Multi-Step Tasks: Tasks requiring 5+ distinct steps or multiple specialized capabilities
• •Cross-Domain Problems: Issues spanning multiple areas (storage, API, testing, documentation)
• •Optimization Opportunities: Tasks that could benefit from parallel or hybrid execution
• •Quality-Critical Work: Projects requiring validation checkpoints and quality gates
• •Resource-Intensive Operations: Large refactors, migrations, or architectural changes
• •Ambiguous Requirements: Tasks needing structured analysis before execution

Available Skills by Category

Quality & Validation Skills

• •rust-code-quality: Comprehensive Rust code review against best practices
• •architecture-validation: Validate implementation vs architecture plans
• •plan-gap-analysis: Implementation gap analysis between plans and code
• •code-quality: General code quality maintenance (formatting, linting)
• •quality-unit-testing: High-quality test writing following best practices

Build & Testing Skills

• •build-compile: Build management and compilation with error handling
• •test-fix: Systematic test debugging and fixing
• •test-runner: Test execution and management

Analysis & Decision-Making Skills

• •analysis-swarm: Multi-perspective code analysis (RYAN, FLASH, SOCRATES)
• •codebase-consolidation: Analyze, consolidate, document codebases
• •debug-troubleshoot: Systematic async Rust debugging

Research Skills

• •web-search-researcher: Web research for modern information
• •context-retrieval: Episodic memory retrieval from learning system

Memory System Skills

• •episode-start: Start learning episodes for task tracking
• •episode-log-steps: Log execution steps during episodes
• •episode-complete: Complete and score episodes
• •memory-mcp: MCP server operations
• •memory-cli-ops: CLI operations for memory system
• •storage-sync: Storage synchronization between Turso and redb

Workflow & Coordination Skills

• •task-decomposition: Break down complex tasks into atomic goals
• •agent-coordination: Coordinate multiple Skills and Agents
• •parallel-execution: Execute independent tasks simultaneously
• •loop-agent: Iterative refinement with convergence detection
• •github-workflows: Diagnose and optimize CI/CD workflows

Meta Skills

• •skill-creator: Create new Claude Code skills
• •feature-implement: Systematic feature implementation workflow

Available Task Agents

Execution Agents

• •feature-implementer: Design, implement, test new features
• •refactorer: Improve code quality, structure, maintainability
• •debugger: Diagnose runtime issues, performance problems

Validation Agents

• •code-reviewer: Review code quality, correctness, standards
• •test-runner: Execute tests, diagnose failures

Meta Agents

• •agent-creator: Create new Task Agents
• •goap-agent: Complex multi-step task planning (recursive)
• •loop-agent: Execute workflows iteratively
• •Explore: Fast codebase exploration and search

Core GOAP Methodology

The GOAP Planning Cycle

1. ANALYZE → Understand goals, constraints, resources
2. DECOMPOSE → Break into atomic tasks with dependencies
3. STRATEGIZE → Choose execution pattern (parallel/sequential/swarm/hybrid/iterative)
4. COORDINATE → Assign tasks to specialized agents
5. EXECUTE → Run with monitoring and quality gates
6. SYNTHESIZE → Aggregate results and validate success

Phase 1: Task Analysis

Initial Assessment

## Task Analysis

**Primary Goal**: [Clear statement of what success looks like]

**Constraints**:
- Time: [Urgent / Normal / Flexible]
- Resources: [Available agents, tools, data]
- Dependencies: [External systems, prerequisites]

**Complexity Level**:
- Simple: Single agent, <3 steps
- Medium: 2-3 agents, some dependencies
- Complex: 4+ agents, mixed execution modes
- Very Complex: Multiple phases, many dependencies

**Quality Requirements**:
- Testing: [Unit / Integration / E2E]
- Standards: [AGENTS.md compliance, formatting, linting]
- Documentation: [API docs, examples, guides]
- Performance: [Speed, memory, scalability]

Context Gathering

1. •Codebase Understanding: Use Explore agent to understand relevant code
2. •Past Patterns: Check if similar tasks have been done before
3. •Available Resources: Identify available agents and their capabilities
4. •Current State: Understand starting conditions and existing implementations

Phase 2: Task Decomposition

Use the task-decomposition skill to break down the goal:

## Task Decomposition: [Task Name]

### Main Goal
[Clear statement of primary objective]

### Sub-Goals
1. [Component 1] - Priority: P0
   - Success Criteria: [How to verify]
   - Dependencies: [Prerequisites]
   - Complexity: [Low/Medium/High]

2. [Component 2] - Priority: P1
   - Success Criteria: [How to verify]
   - Dependencies: [Component 1]
   - Complexity: [Low/Medium/High]

### Atomic Tasks
**Component 1: [Name]**
- Task 1.1: [Action] (Agent: type, Deps: none)
- Task 1.2: [Action] (Agent: type, Deps: 1.1)

### Dependency Graph

Task 1.1 → Task 1.2 → Task 2.1 ↘ Task 1.3 (parallel) → Task 2.2

Key Decomposition Principles

• •Atomic: Each task is indivisible and clear
• •Testable: Can verify completion
• •Independent where possible: Minimize dependencies
• •Assigned: Each task maps to an agent capability

Phase 3: Strategy Selection

Choose execution strategy based on task characteristics. See the GOAP Agent documentation (../agents/goap-agent.md) for detailed execution patterns.

Quick Strategy Guide

Decision Tree

Needs iterative refinement?
  ├─ Yes (until criteria met or converged) → ITERATIVE
  └─ No → Is time critical?
      ├─ Yes → Can tasks run in parallel?
      │   ├─ Yes → PARALLEL
      │   └─ No → SEQUENTIAL (prioritize critical path)
      └─ No → Are tasks similar?
          ├─ Yes (many similar) → SWARM
          ├─ No (mixed) → HYBRID
          └─ Simple linear → SEQUENTIAL

Phase 4: Agent Assignment

Agent Capability Matrix

Assignment Principles

1. •Match agent capabilities to task requirements
2. •Balance workload across agents
3. •Consider agent specialization
4. •Plan for quality validation

Phase-Specific Skill/Agent Recommendations

Phase 1: Research & Analysis

Skills (Parallel):

• •web-search-researcher - Research best practices, modern solutions
• •context-retrieval - Find similar past implementations
• •codebase-consolidation - Understand current architecture

Agents (Parallel):

• •Explore - Fast codebase exploration
• •code-reviewer - Audit current code quality

Use when: Beginning new features, investigating issues, understanding requirements

Phase 2: Decision-Making & Planning

Skills (Sequential):

• •task-decomposition - Break down complex goals
• •analysis-swarm - Multi-perspective architectural decisions (RYAN, FLASH, SOCRATES)

Use when: Multiple valid approaches, significant trade-offs, architectural decisions

Phase 3: Quality Validation (Pre-Implementation)

Skills (Parallel):

• •rust-code-quality - Rust best practices review
• •architecture-validation - Validate vs architectural plans
• •plan-gap-analysis - Verify all requirements covered

Use when: Before major implementation, validating design decisions

Phase 4: Implementation

Agents (Parallel or Sequential):

• •feature-implementer - Build new functionality
• •refactorer - Improve existing code

Skills (Guidance):

• •feature-implement - Feature implementation workflow

Use when: Executing planned work, building features

Phase 5: Testing & Debugging

Skills (Sequential):

• •test-fix - Systematic test debugging
• •quality-unit-testing - High-quality test writing

Agents (Parallel or Sequential):

• •test-runner - Execute test suites
• •debugger - Diagnose runtime issues

Use when: Validating implementations, fixing test failures

Phase 6: Build & CI/CD

Skills (Sequential):

• •build-compile - Build verification and optimization
• •github-workflows - CI/CD pipeline validation

Use when: Preparing for deployment, troubleshooting CI failures

Phase 7: Quality Assurance (Post-Implementation)

Skills (Parallel):

• •rust-code-quality - Final Rust review
• •architecture-validation - Validate vs architecture
• •plan-gap-analysis - Verify completeness

Agents (Parallel):

• •code-reviewer - Final quality check
• •test-runner - Full test suite validation

Use when: Pre-commit, pre-merge, release preparation

Phase 5: Execution Planning

Create the Execution Plan

## Execution Plan: [Task Name]

### Overview
- Strategy: [Parallel/Sequential/Swarm/Hybrid/Iterative]
- Total Tasks: [N]
- Estimated Duration: [Time]
- Quality Gates: [N checkpoints]

### Phase 1: [Phase Name]
**Tasks**:
- Task 1: [Description] (Agent: type)
- Task 2: [Description] (Agent: type)

**Quality Gate**: [Validation criteria]

### Phase 2: [Phase Name]
**Tasks**:
- Task 3: [Description] (Agent: type)

**Quality Gate**: [Validation criteria]

### Overall Success Criteria
- [ ] All tasks complete
- [ ] Quality gates passed
- [ ] Tests passing
- [ ] Documentation updated

### Contingency Plans
- If Phase 1 fails → [Recovery plan]
- If tests fail → [Diagnostic approach]

Phase 6: Coordinated Execution

Parallel Execution

**Launching parallel agents:**
- Agent 1 (feature-implementer) → Task A
- Agent 2 (feature-implementer) → Task B
- Agent 3 (test-runner) → Task C

**Coordination**:
- All agents work simultaneously
- Monitor progress independently
- Aggregate results when all complete

Sequential Execution

**Launching sequential agents:**
Phase 1: Agent 1 (debugger) → Diagnose issue
  ↓ Quality Gate: Root cause identified
Phase 2: Agent 2 (refactorer) → Apply fix
  ↓ Quality Gate: Tests pass
Phase 3: Agent 3 (code-reviewer) → Validate

Monitoring During Execution

• •Track agent progress
• •Monitor for failures
• •Validate intermediate results
• •Adjust plan if needed

Atomic Git Commit Policy

After each successful todo completion, create an atomic git commit:

• •Commit only the changes for that specific todo item
• •Use descriptive commit messages following [module] description format
• •Do NOT commit changes from incomplete todos
• •This ensures incremental, reversible progress tracking
• •Example: feat(storage): add episode creation method

Phase 7: Result Synthesis

Aggregate Results

## Execution Summary: [Task Name]

### ✓ Completed Tasks
- [Task 1]: Success
- [Task 2]: Success

### 📦 Deliverables
- [File/Feature 1]
- [File/Feature 2]

### ✅ Quality Validation
- Tests: [Pass/Fail] ([coverage]%)
- Linting: [Pass/Fail]
- Standards: [Compliant]

### 📊 Performance Metrics
- Duration: [actual vs estimated]
- Efficiency: [parallel speedup if applicable]

### 💡 Recommendations
- [Improvement 1]
- [Improvement 2]

### 🎓 Lessons Learned
- [What worked well]
- [What to improve]

Integration with Self-Learning Memory

GOAP coordination tasks can be tracked as learning episodes to improve future planning decisions.

Starting a GOAP Episode

**Use**: Skill(command="episode-start")

**TaskContext**:
- language: "coordination"
- domain: "goap"
- tags: ["multi-agent", "parallel", "sequential", etc.]

**Description**: "GOAP coordination for [task description]"

Logging GOAP Steps

**Use**: Skill(command="episode-log-steps")

**Log during**:
- Decomposition decisions (how goals were broken down)
- Agent assignments (which agents chosen for which tasks)
- Strategy selection (why parallel vs sequential vs swarm)
- Quality gate results (pass/fail and why)
- Recovery actions (how failures were handled)

Completing a GOAP Episode

**Use**: Skill(command="episode-complete")

**Score based on**:
- Goal achievement (all tasks completed?)
- Efficiency (parallel speedup, resource utilization)
- Quality (all quality gates passed?)
- Adaptability (how well recovered from failures?)

**Patterns extracted**:
- Successful decomposition strategies
- Effective agent assignments
- Optimal execution patterns
- Quality gate effectiveness

Retrieving Past GOAP Context

**Use**: Skill(command="context-retrieval")

**Query for**:
- Similar coordination tasks
- Past parallel/sequential decisions
- Agent assignment patterns
- Quality gate strategies

**Apply learnings**:
- Reuse successful decompositions
- Avoid past mistakes
- Apply proven strategies
- Optimize based on history

Example: Learning-Enabled GOAP

Task: Implement authentication system

Phase 0: Retrieve Context
└─ Skill(command="context-retrieval")
   Query: "authentication implementation coordination"
   → Found: 3 past auth implementations
   → Pattern: Parallel (model + middleware + endpoints) worked well
   → Lesson: Sequential integration after parallel build

Phase 1: Start Episode
└─ Skill(command="episode-start")
   Context: {domain: "goap", tags: ["auth", "parallel"]}

Phase 2-N: Execute with logging
└─ Skill(command="episode-log-steps")
   Log each: decomposition, assignment, quality gate

Phase Final: Complete Episode
└─ Skill(command="episode-complete")
   Score: High (reused successful pattern)
   Pattern: Confirmed parallel → sequential integration strategy

Dynamic Capability Creation

When existing Skills and Agents are insufficient, create new capabilities dynamically.

When to Create New Skills

Create Skill when:

• •Recurring workflow pattern identified
• •Deep domain knowledge needed
• •Reusable methodology discovered
• •No existing Skill covers the domain

Examples:

• •Custom quality standards for your domain
• •Specialized testing workflows
• •Domain-specific architecture patterns
• •Project-specific best practices

How to create:

Use: Skill(command="skill-creator")

Provide:
- Skill name and description
- When to use this skill
- Step-by-step methodology
- Examples and patterns
- Integration points

When to Create New Agents

Create Agent when:

• •New autonomous execution capability needed
• •Specialized tool usage pattern required
• •Cross-cutting concern needs dedicated agent
• •Complex multi-step execution to automate

Examples:

• •Custom deployment agent
• •Specialized migration agent
• •Domain-specific analyzer agent
• •Project-specific workflow agent

How to create:

Use: Task(subagent_type="agent-creator", ...)

Or use: Skill(command="skill-creator") for agent definition

Provide:
- Agent purpose and capabilities
- Tools the agent needs
- Input/output specification
- Success criteria

Update GOAP Knowledge

After creating new capabilities:

1. •

   Document in GOAP:

   • •Add to Skills or Agents list
   • •Update capability matrix
   • •Add to phase-specific recommendations
2. •

   Test the capability:

   • •Use in real scenario
   • •Validate effectiveness
   • •Refine as needed
3. •

   Share the pattern:

   • •Document in project
   • •Add examples
   • •Enable reuse

Example: Creating Custom Capability

Problem: Need specialized security audit for authentication code

Step 1: Identify gap
→ No existing Skill covers auth security audit specifically

Step 2: Create Skill
└─ Skill(command="skill-creator")
   Name: "auth-security-audit"
   Purpose: "Audit authentication code for security vulnerabilities"
   Methodology: [OWASP auth checklist, crypto review, token validation, ...]

Step 3: Integrate into GOAP
→ Add to Quality & Validation Skills
→ Add to Phase 3 and Phase 7 recommendations
→ Document in project CLAUDE.md

Step 4: Use in workflow
└─ Phase 3: Skill(command="auth-security-audit")
   → Validates auth design before implementation

Common GOAP Patterns

Pattern 1: Research → Decide → Implement → Validate (Full Stack)

Task: Implement complex feature with architectural impact

Phase 0: Retrieve Context [Skills]
├─ Skill(command="context-retrieval")
│  Query: "similar feature implementations"
│  → Apply past learnings
└─ Skill(command="episode-start")
   → Start tracking this coordination

Phase 1: Research [Parallel Skills + Agents]
├─ Skill(command="web-search-researcher")
│  → Research modern best practices
├─ Skill(command="codebase-consolidation")
│  → Understand current architecture
└─ Task(subagent_type="Explore")
   → Fast codebase exploration
Quality Gate: Architecture and requirements clear

Phase 2: Decision [Skill]
└─ Skill(command="analysis-swarm")
   → Multi-perspective architectural decision (RYAN, FLASH, SOCRATES)
   → Evaluate trade-offs, choose approach
Quality Gate: Architecture approved

Phase 3: Pre-Implementation Validation [Parallel Skills]
├─ Skill(command="architecture-validation")
│  → Validate design vs plans
├─ Skill(command="plan-gap-analysis")
│  → Ensure complete requirements coverage
└─ Skill(command="rust-code-quality")
   → Review design for Rust best practices
Quality Gate: Design validated

Phase 4: Implementation [Parallel Agents]
├─ Task(subagent_type="feature-implementer")
│  Prompt: "Implement Module A"
├─ Task(subagent_type="feature-implementer")
│  Prompt: "Implement Module B"
└─ Task(subagent_type="feature-implementer")
   Prompt: "Implement Module C"
Quality Gate: All modules implemented

Phase 5: Testing [Skills + Agents]
├─ Task(subagent_type="test-runner")
│  → Execute all tests
├─ Skill(command="test-fix")
│  → Fix any failing tests systematically
└─ Skill(command="quality-unit-testing")
   → Ensure high-quality tests
Quality Gate: All tests passing

Phase 6: Quality Validation [Parallel Skills + Agents]
├─ Skill(command="rust-code-quality")
│  → Final Rust review
├─ Skill(command="architecture-validation")
│  → Validate vs architecture
├─ Skill(command="plan-gap-analysis")
│  → Verify completeness
└─ Task(subagent_type="code-reviewer")
   → Final quality check
Quality Gate: Quality standards met

Phase 7: Build & CI [Skills]
├─ Skill(command="build-compile")
│  → Build verification
└─ Skill(command="github-workflows")
   → CI validation
Quality Gate: Ready for merge

Phase 8: Learning [Skills]
└─ Skill(command="episode-complete")
   Score: High (all phases successful)
   Patterns: Document successful strategies

Pattern 2: Investigate → Diagnose → Fix → Verify

Phase 1: Investigate
  - debugger → Reproduce issue
  - Quality Gate: Issue reproduced

Phase 2: Diagnose
  - debugger → Root cause analysis
  - Quality Gate: Root cause identified

Phase 3: Fix
  - refactorer → Apply fix
  - Quality Gate: Fix implemented

Phase 4: Verify
  - test-runner → Regression tests
  - Quality Gate: Tests pass

Pattern 3: Audit → Improve → Validate

Phase 1: Audit
  - code-reviewer → Quality audit
  - Quality Gate: Issues identified

Phase 2 (Swarm): Improve
  - Multiple refactorer agents
  - Work queue: [issue list]
  - Quality Gate: All issues addressed

Phase 3: Validate
  - test-runner → Full test suite
  - code-reviewer → Final check
  - Quality Gate: Quality targets met

Error Handling & Recovery

Agent Failure Recovery

**If agent fails:**
1. Log failure reason
2. Check quality gate status
3. Options:
   - Retry same agent (transient error)
   - Assign to different agent (agent issue)
   - Modify task (requirements issue)
   - Escalate to user (blocking issue)

Quality Gate Failure

**If quality gate fails:**
1. Identify failing criteria
2. Diagnose root cause
3. Options:
   - Re-run previous phase with fixes
   - Adjust quality criteria (if appropriate)
   - Change strategy (e.g., parallel → sequential for debugging)

Blocked Dependencies

**If dependency blocks progress:**
1. Identify blocking task
2. Prioritize unblocking
3. Options:
   - Execute dependency first (re-order)
   - Remove dependency (refactor plan)
   - Parallel work on independent tasks

Best Practices

DO:

✓ Break tasks into atomic, testable units ✓ Define clear quality gates between phases ✓ Match agent capabilities to task requirements ✓ Monitor progress and validate incrementally ✓ Document decisions and rationale ✓ Learn from execution for future planning ✓ Use parallel execution where safe ✓ Validate dependencies before execution ✓ Create plan to fix all pre-existing issues ✓ Never skip lint - always run and resolve ✓ Always implement changes and verify they work

DON'T:

✗ Create monolithic tasks (break them down) ✗ Skip quality gates (leads to cascading failures) ✗ Assume tasks are independent (verify carefully) ✗ Ignore agent failures (address immediately) ✗ Over-complicate simple tasks (use sequential) ✗ Under-estimate coordination overhead ✗ Forget to aggregate and synthesize results ✗ Leave pre-existing issues unfixed ✗ Skip lint checks or warnings

Integration with Other Skills

• •task-decomposition: Use for Phase 2 (breaking down complex goals)
• •agent-coordination: Use for Phase 6 (coordinating multiple agents)
• •parallel-execution: Use for parallel strategy implementation
• •loop-agent: Use for iterative refinement strategy implementation
• •All specialized agents (feature-implementer, debugger, test-runner, etc.)

Quick Example

Task: Implement authentication system

## GOAP Plan

### Phase 1: Analysis (Sequential)
- goap-agent → Define requirements
- Quality Gate: Requirements clear

### Phase 2: Implementation (Parallel)
- Agent A → User model + database
- Agent B → Auth middleware
- Agent C → API endpoints
- Quality Gate: All components implemented

### Phase 3: Integration (Sequential)
- feature-implementer → Wire components together
- test-runner → Integration tests
- Quality Gate: Tests pass

### Phase 4: Validation (Sequential)
- code-reviewer → Security review
- Quality Gate: Approved for deployment

Success Metrics

Planning Quality

• •Clear decomposition with measurable tasks
• •Realistic time estimates
• •Appropriate strategy selection
• •Well-defined quality gates

Execution Quality

• •Tasks completed as planned
• •Quality gates passed
• •Minimal re-work required
• •Efficient resource utilization

Learning

• •Document what worked well
• •Identify improvement areas
• •Update patterns for future use
• •Share knowledge with team

Advanced Topics

Dynamic Re-Planning

If during execution:

• •Dependencies change
• •Requirements clarified
• •Blockers discovered
• •Performance issues found

Then:

1. •Pause execution
2. •Re-analyze with new information
3. •Adjust plan (tasks, dependencies, strategy)
4. •Resume with updated plan

Optimization Techniques

• •Critical path optimization: Parallelize non-critical-path tasks
• •Resource pooling: Share agents across similar tasks
• •Incremental delivery: Complete and validate phases incrementally
• •Adaptive strategy: Switch strategies based on progress

Summary

GOAP enables systematic planning and execution of complex tasks through:

1. •Structured Analysis: Understand goals, constraints, and context
2. •Intelligent Decomposition: Break into atomic, testable tasks
3. •Strategic Execution: Choose optimal execution pattern
4. •Quality Assurance: Validate at checkpoints
5. •Coordinated Agents: Leverage specialized capabilities
6. •Continuous Learning: Improve from each execution

Use GOAP when facing complex, multi-step challenges requiring coordination, optimization, and quality assurance.