Build Command

Execute builds through sequential-thinking planning and complete agent chain workflow.

Use when: You know what you want to build. For exploration/planning, use /architect instead.

---

Entry Handling

When /build is invoked, ALWAYS proceed to Phase 1 regardless of prompt content.

---

Phase 1: Planning (Sequential Thinking)

Use: sequential-thinking MCP tool

Plan the approach using sequential-thinking MCP tool:

1. Analyze the prompt to understand scope and requirements
2. Identify target repositories and reference patterns
3. Outline high-level implementation strategy
4. Note potential risks and dependencies

This creates the mental model for the full workflow.

---

Phase 2: Analysis

Follow: /skill analysis-phase

Spawn scope-check agent to identify target and reference repositories. Use scope-check agent results directly. Proceed with identified targets immediately.

Output: scope: { targets[], references[] }

---

Phase 3: Requirements

Follow: /skill requirements-phase

EA uses AskUserQuestion until 100% understanding, then creates PRD:

Enterprise Architect (EA) → AskUserQuestion loop until fully understood
    → Creates PRD (prd-creation skill)
    → Reviewer → validates PRD (document-validation)
    → (fix if needed, loop back to EA)
    → Business Analyst (BA) → extracts epics and story outlines

Skills used internally:

• /skill prd-creation (EA creates PRD)
• /skill document-validation (Reviewer validates)

Output: PRD + Epics + Story outlines (NO HITL - continues to Design)

Enforcement Gate (MANDATORY):

1. Call checkPhaseRequirements(projectFolder, 3) from /skill state-management
2. Verify prd.md exists in project folder
3. Spawn reviewer agent for PRD validation
4. On FAIL: Loop back to EA for PRD creation/fixes
5. On PASS: Continue to Phase 4 (Design)

---

Phase 4: Design

Follow: /skill design-phase

Agents use AskUserQuestion for any clarifications, then create all docs:

Architect → annotates PRD with implementation details
    → BA → creates story outlines
    → PM → creates execution plan (execution-plan-creation)
    → Reviewer → validates execution plan (document-validation)
    → (fix if needed, loop back to PM)
    → BA → fills story details (user-story-creation)
    → Architect → adds Architecture sections
    → Reviewer → validates stories (document-validation)
    → (fix if needed, loop back to BA)

Skills used internally:

• /skill execution-plan-creation (PM creates plan)
• /skill user-story-creation (BA fills stories)
• /skill document-validation (Reviewer validates)

Output: Execution plan + Complete user stories (NO HITL - continues to Approval)

Enforcement Gate - Execution Plan (MANDATORY):

1. Call checkPhaseRequirements(projectFolder, 5) from /skill state-management
2. Verify execution-plan.md exists in project folder
3. Spawn reviewer agent for plan validation
4. On FAIL: Loop back to PM for execution plan creation/fixes
5. On PASS: Continue to story creation

Enforcement Gate - User Stories (MANDATORY):

1. Call checkPhaseRequirements(projectFolder, 6) from /skill state-management
2. Verify user-stories/ folder is populated (at least 1 file)
3. Spawn reviewer agent for stories validation
4. On FAIL: Loop back to BA for story creation/fixes
5. On PASS: Continue to Phase 5 (HITL Approval)

---

Phase 5: HITL Approval

Follow: /skill hitl-approval

SINGLE approval gate for all documentation:

Present to user:

1. PRD summary (from Phase 3)
2. Execution plan summary (waves, dependencies)
3. User stories summary (count, complexity)

Use AskUserQuestion:

• APPROVE → Continue to Execution
• REJECT → Collect feedback, return to appropriate phase (Requirements or Design)

---

Phase 6: Execution

Follow: /skill execution-phase

TDD-paired wave-based execution:

1. Load execution plan (waves with dependencies)
2. For each wave:
   • Update workflow state
   • Run /compact to free context
   • Spawn tester agent (writes tests from AC)
   • Spawn implementation agent (passes tests)
   • Production verification (tests pass, AC marked)
3. Waves execute continuously with state persistence

Wave Checkpoint Flow:

1. Wave N agents complete their stories
2. Quick log - /skill workflow-postmortem mode=log (30 seconds max)
3. Compact context - Run /compact to manage context window
4. Proceed to Wave N+1

---

Phase 7: Validation

Follow: /skill validation-phase

1. Verify all acceptance criteria marked complete
2. Run build/lint/test (pnpm build, pnpm lint, pnpm test)
3. Run code quality checks scaled by change size

On failure: Return to Execution for fixes.

---

Phase 8: Standards Audit

Follow: /skill structure-check, /skill dry-check

Three checks (PARALLEL):

1. Structure Check: Validate files in correct locations per domain
2. DRY Check: Scan new code against shared libraries
3. Config Audit: For modified config files, spawn appropriate config agents from agents/config/ in audit mode

On failure: Return to Execution for fixes, re-run Validation, re-run Standards Audit, loop until pass.

---

Phase 9: Final Workflow Postmortem

Follow: /skill workflow-postmortem mode=summary

Run /skill workflow-postmortem mode=summary to generate final summary. This reads the accumulated wave logs from docs/epics/{project}/post-mortem.md and presents a summary to the user.

Output: Summary of issues logged across waves (count by category, patterns identified).

---

Phase 10: Report

Follow: /skill report-phase

BA + PM produce final build report with:

• Summary of changes
• Files modified
• Tests added
• Standards audit results
• Workflow postmortem
• Next steps

---

Examples

/build "add logging to service"
→ Planning → Analysis → Requirements (EA asks questions → PRD → Reviewer) → Design (Architect → BA → PM → Reviewer) → HITL (approve all docs) → Execution → Validation → Standards Audit → Postmortem → Report

/build "refactor auth module"
→ Planning → Analysis → Requirements → Design → HITL → Execution (waves) → Validation → Standards Audit → Postmortem → Report

/build "multi-tenant SaaS"
→ Planning → Analysis → Requirements → Design → HITL → Wave1 → Wave2 → ... → Validation → Standards Audit → Summary → Report

# Wave checkpoint flow (no HITL between waves):
Wave1: Task(spawn tester) → Task(spawn coder) → /skill workflow-postmortem mode=log → /compact
Wave2: Task(spawn tester) → Task(spawn coder) → /skill workflow-postmortem mode=log → /compact
End: /skill workflow-postmortem mode=summary → Report

# Spawning agents with Task tool:
Task: subagent_type="core-claude-plugin:generic:tester" prompt="Execute /skill tdd-execution for story X"
→ (tester creates test file)
Task: subagent_type="core-claude-plugin:generic:coder" prompt="Implement against tests from story X"
→ (coder implements feature)

---

Enforcement

1. Frame all instructions positively. State what to do.
2. ALWAYS run sequential-thinking MCP tool first to plan approach.
3. ALWAYS run full workflow: Planning → Analysis → Requirements → Design → HITL → Execution → Validation → Standards Audit → Postmortem → Report
4. Requirements phase: EA uses AskUserQuestion until fully understood, then creates PRD → Reviewer validates → BA extracts stories (NO HITL)
5. Design phase: Architect → BA → PM → Reviewer validates all docs (NO HITL - continues to single HITL gate)
6. SINGLE HITL GATE: After all docs created (PRD, execution plan, stories), present summary and get approval
7. TDD execution: ALWAYS run tester BEFORE coder per story.
8. ALWAYS use Task tool to spawn tester agent first, then coder agent for implementation work.
9. ALWAYS spawn agents for implementation - the orchestrator tracks progress while agents write code.
10. Standards Audit: ALWAYS run AFTER Validation passes.
11. ALWAYS run build/lint/test after execution.
12. WAVE CHECKPOINT TIMING: ALWAYS run /skill workflow-postmortem mode=log BEFORE compact at each wave checkpoint.
13. POSTMORTEM LOG ACCUMULATION: Each wave log appends to docs/epics/{project}/post-mortem.md.
14. ALWAYS run /skill workflow-postmortem mode=summary AFTER Standards Audit, BEFORE Report.
15. ALWAYS produce final report with workflow postmortem section.
16. When files modified, spawn agent: subagent_type="general-purpose" with prompt "Execute /skill repomix-cache-refresh"
17. Use AskUserQuestion tool for clarifications during Requirements and Design phases.
18. Use /architect command for innovation or vibe check workflows.
19. Git operations are outside workflow scope. Changes remain uncommitted for user to handle.
20. PHASE GATE - PRD: ALWAYS call checkPhaseRequirements(projectFolder, 3) after Requirements phase. Verify prd.md exists before continuing to Design.
21. PHASE GATE - EXECUTION PLAN: ALWAYS call checkPhaseRequirements(projectFolder, 5) after execution plan creation. Verify execution-plan.md exists before story creation.
22. PHASE GATE - USER STORIES: ALWAYS call checkPhaseRequirements(projectFolder, 6) after story creation. Verify user-stories/ folder is populated before HITL Approval.
23. GATE FAILURE HANDLING: On gate failure, loop back to the producing agent (EA for PRD, PM for plan, BA for stories) and retry until artifact exists.
24. GATE VALIDATION: Use /skill state-management for all checkPhaseRequirements calls.