recipe-build

name	recipe-build
description	Execute decomposed backend tasks in autonomous execution mode using task-executor and quality-fixer.

Required Skills [LOAD BEFORE EXECUTION]

1. [LOAD IF NOT ACTIVE] coding-rules — coding standards
2. [LOAD IF NOT ACTIVE] testing — test strategy and quality gates
3. [LOAD IF NOT ACTIVE] ai-development-guide — AI development patterns
4. [LOAD IF NOT ACTIVE] subagents-orchestration-guide — agent coordination and workflow flows

Spawn rule: every spawn_agent call MUST pass fork_turns="none" or fork_context=false for context isolation.

Orchestrator Definition

Core Identity: "I am not a worker. I am an orchestrator." (see subagents-orchestration-guide skill)

Execution Protocol:

1. Spawn agents for all work -- your role is to invoke sub-agents, pass data between them, and report results
2. Follow the 4-step task cycle exactly: task-executor -> escalation check -> quality-fixer -> commit
3. Enter autonomous mode when user provides execution instruction with existing task files -- this IS the batch approval
4. Scope: Complete when all tasks are committed or escalation occurs

CRITICAL: MUST run quality-fixer before every commit. ENFORCEMENT: Commits without quality-fixer approval are invalid and MUST be reverted.

Work plan: $ARGUMENTS

Pre-execution Prerequisites

Implementation Readiness Check

Before task processing, locate the work plan to gate against.

Resolution rule:

1. If $ARGUMENTS contains a work plan path, use that exact file and derive {plan-name} from its basename. This takes precedence over task-file mtimes.
2. If $ARGUMENTS is empty, list task files in docs/plans/tasks/ matching the single-layer pattern {plan-name}-task-*.md.
3. Exclude *-task-prep-*.md, _overview-*.md, *-phase*-completion.md, review-fixes-*.md, and integration-tests-*-task-*.md.
4. If matching task files exist, infer {plan-name} from the most recent matching task file and use docs/plans/{plan-name}.md.
5. If no matching task files exist, use the most recent non-template work plan in docs/plans/.

Read the work plan header and apply the Implementation Readiness Marker Contract from subagents-orchestration-guide.

Consumed Task Set

Compute the Consumed Task Set for this run: task files in docs/plans/tasks/ matching {plan-name}-task-*.md, excluding *-task-prep-*.md, _overview-*.md, *-phase*-completion.md, review-fixes-*.md, and integration-tests-*-task-*.md.

Every subsequent reference to task files in this recipe uses this set, not an unrestricted docs/plans/tasks/*.md scan.

Task Generation Decision Flow

Analyze task file existence state and determine the action required:

State	Criteria	Next Action
Tasks exist	Consumed Task Set is non-empty	User's execution instruction serves as batch approval -> Enter autonomous execution immediately
No tasks + plan exists + reviewed plan	Consumed Task Set is empty and WorkPlan Review records Status: approved, Conditions: none	Confirm with user -> spawn task-decomposer
No tasks + small simplified plan	Consumed Task Set is empty, plan exists, and the plan references no Design Doc	Confirm with user -> spawn task-decomposer
No tasks + plan exists + unreviewed plan	Consumed Task Set is empty, the plan references a Design Doc, and WorkPlan Review is absent, pending, conditional, or not approved	Run work plan review, then confirm with user -> spawn task-decomposer
Neither exists	No plan or task files	Error: Prerequisites not met

Task Decomposition Phase (Conditional)

When task files don't exist, the plan references a Design Doc, and the WorkPlan Review section is absent, pending, conditional, or not approved:

1. Work Plan Review

Spawn document-reviewer agent: "Review the work plan before task decomposition. doc_type: WorkPlan. target: docs/plans/[plan-name].md. mode: composite. Review semantic traceability to the Design Doc, early verification placement, real-boundary verification coverage, Proof Strategy, Failure Mode Checklist, Review Scope, and Quality Assurance coverage."

Branch on verdict.decision:

• approved -> spawn work-planner in update mode once to record Status: approved and Conditions: none in WorkPlan Review, then continue to user confirmation
• approved_with_conditions -> stop before task decomposition and report that the work plan needs update via recipe-plan
• needs_revision -> stop before task decomposition and report that the work plan needs update via recipe-plan
• rejected -> stop before task decomposition and present the blocking findings to the user

When task files don't exist and the WorkPlan Review section records Status: approved and Conditions: none, skip Work Plan Review and continue to user confirmation.

2. User Confirmation

No task files found.
Work plan: docs/plans/[plan-name].md

Generate tasks from the work plan? (y/n):

3. Task Decomposition (if approved)

Spawn task-decomposer agent: "Read work plan at docs/plans/[plan-name].md and decompose into atomic tasks. Output: Individual task files in docs/plans/tasks/. Granularity: 1 task = 1 commit = independently executable."

4. Verify Generation

Recompute the Consumed Task Set and verify it is non-empty.

Pre-execution Checklist

• Confirmed task files exist in docs/plans/tasks/
• Identified task execution order (dependencies)
• Environment check: Can I execute per-task commit cycle?
  • If commit capability unavailable -> Escalate before autonomous mode
  • Other environments (tests, quality tools) -> Subagents will escalate

Task Execution Cycle (4-Step Cycle)

MANDATORY EXECUTION CYCLE: task-executor -> escalation check -> quality-fixer -> commit

For EACH task, YOU MUST:

1. Register tasks: Register work steps. Always include: first "Confirm skill constraints", final "Verify skill fidelity"
2. Spawn task-executor agent: "Execute the task implementation for [task-file-path]"
3. CHECK task-executor response:
   • status: "escalation_needed" or "blocked" -> STOP and escalate to user
   • requiresTestReview is true -> Spawn integration-test-reviewer agent: "Review integration tests in [test-files]"
     • needs_revision -> Return to step 2 with requiredFixes
     • approved -> Proceed to step 4
   • readyForQualityCheck: true -> Proceed to step 4
4. Spawn quality-fixer agent: "Execute all quality checks and fixes. Task file: [task-file-path]. The task file path above is also the task_file input. Read its Quality Assurance Mechanisms section as supplementary quality-check hints. filesModified: [task-executor response filesModified]. Use these files as the stub-detection scope."
5. CHECK quality-fixer response:
   • status: "stub_detected" -> Return to step 2 with stubFindings
   • status: "blocked" -> STOP and escalate to user
   • status: "approved" -> Proceed to step 6
6. COMMIT on approval: After status: "approved" from quality-fixer -> Execute git commit

CRITICAL: MUST monitor ALL structured responses WITHOUT EXCEPTION and ENSURE every quality gate is passed. ENFORCEMENT: Proceeding past a failed quality gate invalidates all subsequent work.

Sub-agent Invocation Constraints

MANDATORY suffix for ALL sub-agent prompts:

[SYSTEM CONSTRAINT]
This agent operates within build skill scope. Use the task file as the primary instruction source. Use the active Design Doc or work plan only as supporting context when the task file references them. Constraints explicitly passed in this prompt by the orchestrator take precedence over supporting context. The agent's own role contract and required quality rules remain in force.

Autonomous sub-agents require scope constraints for stable execution. MUST append this constraint to every sub-agent prompt. ENFORCEMENT: Sub-agent prompts missing the constraint suffix MUST be re-issued with the constraint appended.

VERIFY approval status before proceeding. Once confirmed, INITIATE autonomous execution mode.

Post-Implementation Verification (After All Tasks Complete)

After all task cycles finish, collect all filesModified from every task-executor response (deduplicated), then run both verification agents before the completion report:

1. Spawn code-verifier agent: "Verify implementation consistency against the Design Doc. doc_type: design-doc. document_path: [path]. code_paths: [collected filesModified list]. Work Plan Review Scope: [Review Scope value from the active work plan, used only to confirm the collected file set is complete]."
2. Spawn security-reviewer agent: "Design Doc: [path]. Implementation files: [collected filesModified list]. Review security compliance."
3. Consolidate results:
   • code-verifier passes when summary.status is consistent or mostly_consistent
   • code-verifier fails when summary.status is needs_review or inconsistent
   • security-reviewer passes when status is approved or approved_with_notes
   • security-reviewer fails when status is needs_revision
   • security-reviewer blocked -> Escalate to user
4. If either verifier fails:
   • Create a single fix task covering verifier discrepancies and security requiredFixes
   • Spawn task-executor with that consolidated task
   • Spawn quality-fixer
   • Re-run only the verifier(s) that failed
   • Maximum retry count is 1 verification fix cycle; if any failed verifier still fails after re-run, escalate to the user
5. If both verifiers pass -> Proceed to completion report

Final Cleanup

Before the completion report, delete only these files for the current {plan-name}:

• Every file in the Consumed Task Set
• docs/plans/tasks/{plan-name}-phase*-completion.md
• docs/plans/tasks/_overview-{plan-name}.md

Preserve the work plan itself.

If cleanup fails, report the failed path but do not invalidate completed implementation work.

[STOP — BLOCKING] Upon detecting ANY requirement changes, halt execution immediately. CANNOT proceed until user explicitly confirms the change scope.

Completion Criteria

• Task files verified in docs/plans/tasks/
• Task execution order identified with dependencies
• Environment check completed (commit capability confirmed)
• All tasks executed through 4-step cycle (task-executor -> check -> quality-fixer -> commit)
• System constraint suffix appended to all sub-agent prompts
• All quality gates passed
• All tasks committed or escalation completed

Output Example

Backend implementation phase completed.

• Task decomposition: Generated under docs/plans/tasks/
• Implemented tasks: [number] tasks
• Quality checks: All passed
• Commits: [number] commits created