Execution Plan Creation Skill

Purpose: Guide execution plan creation following the standard template Trigger: When PM agent creates or updates an execution plan document Output: Well-structured execution plan with waves, dependencies, and agent assignments

---

Template Reference

ALWAYS read before creating: templates/execution-plan-template.md

Contains: frontmatter fields, summary metrics, wave structure, story index, verification gates, agent assignments, risk/rollback sections.

---

Workflow

1. Read the template:

   • Load templates/execution-plan-template.md
   • Note all required frontmatter fields and section structure

2. Gather inputs:

   • Read approved PRD for project context
   • Read all user stories from user-stories/ folder
   • Identify story dependencies and complexity scores

3. Analyze dependencies:

   • Build dependency graph from story prerequisites
   • Validate DAG structure (detect cycles)
   • Identify critical path through dependencies

4. Organize into waves:

   • Wave 1: Stories with no dependencies
   • Subsequent waves: Stories whose dependencies are satisfied
   • Group parallel-executable stories within each wave

5. Assign agents:

   • Match story requirements to agent capabilities
   • Use full subagent_type paths (e.g., core-claude-plugin:generic:coder)
   • Reference /skill agent-selection for assignment guidance

6. Define verification gates:

   • Add build/lint/test commands per wave
   • Include integration checks between waves
   • Define rollback triggers

7. Validate structure:

   • Run validation checklist (see below)
   • Fix any missing or malformed sections

8. Save execution plan:

   • Write to {projectFolder}/execution-plan.md
   • Ensure frontmatter is complete

9. Validation gate:

   • Spawn reviewer for mandatory plan validation
   • On PASS: Continue to execution phase
   • On FAIL: Loop back with issues for revision

---

Validation Gate

After execution plan is written, spawn reviewer for mandatory validation.

Spawn: core-claude-plugin:generic:reviewer Input: Plan path, validation type execution-plan Output: {result: "PASS"|"FAIL", issues: []}

Result	Action
PASS	Continue to next phase (wave execution)
FAIL	Return issues to project-manager, retry from step 7

Validation Scope:

• Frontmatter completeness
• Wave dependency integrity (valid DAG)
• Agent assignment validity
• Verification gates defined per wave

---

Wave Organization Rules

Rule	Guidance
Wave 1 = foundation	Stories with no dependencies, can run in parallel
Minimize wave count	Group stories by dependency satisfaction
Parallel within wave	Stories in same wave execute concurrently
Sequential between waves	Wave N+1 waits for Wave N verification gate
Critical path awareness	Identify longest dependency chain

---

Agent Assignment Guidelines

Reference /skill agent-selection for complete agent listing. Common assignments:

Story Type	Agent	subagent_type
General implementation	coder	core-claude-plugin:generic:coder
Backend/API	backend-dev	core-claude-plugin:generic:backend-dev
Database/Prisma	prisma-database-agent	core-claude-plugin:domain:database:prisma-database-agent
React Components	react-component-agent	core-claude-plugin:domain:frontend:react-component-agent
Unit Tests	unit-test-agent	core-claude-plugin:domain:testing:unit-test-agent
Integration Tests	integration-test-agent	core-claude-plugin:domain:testing:integration-test-agent
Config files	(specific config agent)	core-claude-plugin:config:{category}:{agent-name}
Code review	reviewer	core-claude-plugin:generic:reviewer
Agent/Skill creation	agent-author	core-claude-plugin:generic:agent-author

Full subagent_type format: core-claude-plugin:{category}:{subcategory}:{agent-name}

---

Validation Checklist

Frontmatter (all required)

• project_id, title, status, owner present
• total_stories, total_complexity, total_waves accurate
• created, updated dates (YYYY-MM-DD)

Structure

• Summary table with metrics
• User Stories Index with all stories
• Wave sections with verification gates
• Agent Assignments table complete

Wave Integrity

• Every story in exactly one wave
• Wave 1 = no dependencies (parallel)
• Subsequent waves depend only on prior waves
• No circular dependencies (valid DAG)

Agent Assignments

• Every story has assigned agent
• Full subagent_type format used
• Agent matches story type

---

Example

Input: PRD with 6 user stories

Wave Assignment:

• Wave 1: msm-auth-001, msm-auth-002 (no deps, parallel)
• Wave 2: msm-auth-003, msm-auth-004 (depend on Wave 1)
• Wave 3: msm-auth-005, msm-auth-006 (depend on Wave 2)

Output: docs/epics/msm007-auth-api/execution-plan.md with 3 waves, agent assignments, verification gates

---

Integration

Called by: planning-phase, /architect, /build, /ms commands Calls: Read (template, PRD, stories), Write (save plan), Task (reviewer validation gate) Spawns: core-claude-plugin:generic:reviewer (mandatory validation gate) References: /skill agent-selection, /skill user-story-template, /skill execution-phase

Previous: Story extraction | Next: Wave execution (on PASS) | Retry (on FAIL)