mcpField
G
goap-agent

goap-agent

github.com/d-o-hubAgent workflows1336 installsUpdated today
Repository

About goap-agent

goap-agent is a planning skill for Claude that structures complex, multi-step work through goal decomposition, execution strategies, and specialized-agent coordination. It also includes guidance for quality gates and checking architectural decisions.

Key features

  1. 01Goal-oriented task planning
  2. 02Atomic task decomposition
  3. 03Parallel and sequential strategies
  4. 04Specialized-agent coordination
  5. 05Quality gates and validation
  6. 06ADR-informed execution plans

Use cases

  1. 01Planning tasks with five or more steps
  2. 02Coordinating work across domains
  3. 03Managing large refactors or architecture changes

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/d-o-hub/rust-self-learning-memory --skill goap-agent

After installing, type/goap-agentin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

agents.md
examples.md
execution-strategies.md
methodology.md
patterns.md
skills.md

Details

From the project README

GOAP Agent Skill

Goal-Oriented Action Planning for complex multi-step tasks with intelligent planning and multi-agent coordination.

Quick Reference

When to Use

  • Complex multi-step tasks (5+ distinct steps)
  • Cross-domain problems (storage, API, testing, documentation)
  • Tasks requiring parallel/sequential execution
  • Quality-critical work with validation checkpoints
  • Large refactors or architectural changes

CRITICAL: Skills vs Task Agents

Skills (via Skill tool): Instruction sets that guide Claude directly Agents (via Task tool): Autonomous sub-processes that execute tasks

Example:

  • WRONG: Task(subagent_type="code-quality", ...) → ERROR!
  • CORRECT: Skill(command="code-quality") → SUCCESS

See skills.md for complete skills list and agents.md for agent capabilities.

Core Process

  1. ANALYZE → Understand goals, constraints, resources
    • Check ADRs: Read relevant ADRs from plans/adr/ before planning
  2. DECOMPOSE → Break into atomic tasks with dependencies
    • Task Decomposition: Create goal hierarchy with success criteria (see patterns.md)
    • Dependency Types: Sequential (A→B), Parallel (A─┐B─┐), Converging (A─┐B─┼─>D)
  3. STRATEGIZE → Choose execution pattern
  4. COORDINATE → Assign to specialized agents
  5. EXECUTE → Run with monitoring and quality gates
  6. SYNTHESIZE → Aggregate results and validate success

PR Monitoring Guardrail

  • During EXECUTE/SYNTHESIZE for PR work, always verify statusCheckRollup on the latest head SHA.
  • Treat an empty required-check rollup as a blocker and document it in plans/STATUS/VALIDATION_LATEST.md.
  • Avoid adding plans-only follow-up commits until remediation checks are attached.

See methodology.md for detailed phase-by-phase guidance and patterns.md for common execution patterns.

ADR Integration Workflow

MANDATORY: Always check ADRs in plans/adr/ before creating execution plans:

Step 1: ADR Discovery
# List all ADRs to identify relevant ones
ls plans/adr/ADR-*.md
Step 2: Read Relevant ADRs
  • Search for ADRs related to your task domain
  • Note architectural decisions and constraints
  • Check ADR status (Accepted/Implemented vs Deprecated)
Step 3: Incorporate into Planning
  • Use ADR constraints when decomposing tasks
  • Reference ADRs in execution plans
  • Ensure compliance with architectural decisions
Step 4: Update Progress in plans/
  • Create/update execution plan files in plans/
  • Document progress, blockers, and decisions
  • Link to relevant ADRs in plan files

Comments

Markdown supported