/build-plan
Source: ~/.claude/skills/build-plan/SKILL.md
name: build-plan description: Execute an approved plan using TaskList with builder/validator teams. Run after /plan-with-team. argument-hint: "[path to plan file, or leave empty for latest]"
Build Plan
Execute a plan using the TaskList system with builder/validator teams.
Instructions
You are executing an approved plan by creating tasks and spawning agents.
Step 1: Load Plan
If path provided, read that file. Otherwise, find most recent .md in ~/system/specs/.
Extract:
- Team members (builders and validators)
- Tasks with acceptance criteria
- Dependencies (blockedBy)
Step 2: Create Tasks
For each task in the plan:
TaskCreate({
subject: "[Task title from plan]",
description: "[Full description + acceptance criteria]",
activeForm: "[Present tense action, e.g., 'Building auth module']"
})
Step 3: Set Dependencies
For each validator task, block it by its builder:
TaskUpdate({
taskId: "[validator-id]",
addBlockedBy: ["[builder-id]"]
})
Step 4: Execute Builders (Parallel)
For all unblocked builder tasks, spawn agents IN PARALLEL using Task tool:
Task({
subagent_type: "general-purpose",
description: "Build [component]",
prompt: `You are a BUILDER agent. Read ~/.claude/agents/builder.md.
Your task ID: [id]
Task: [description]
Acceptance criteria: [criteria]
When done, report what you built.`
})
Step 5: Update Builder Status
When builders complete:
TaskUpdate({ taskId: "[id]", status: "completed" })
Step 6: Execute Validators (Parallel)
Spawn validator agents for completed builders:
Task({
subagent_type: "general-purpose",
description: "Validate [component]",
prompt: `You are a VALIDATOR agent. Read ~/.claude/agents/validator.md.
Your task ID: [id]
Task: [description]
Check: [acceptance criteria]
READ-ONLY. Report PASS or FAIL.`
})
Step 7: Update & Report
When all tasks complete, show:
# Build Complete
## Summary
- Total tasks: X
- Passed: Y
- Failed: Z
## Tasks
- [x] Task 1 — PASS
- [x] Task 2 — PASS
- [ ] Task 3 — FAIL: [reason]
## Files Created/Modified
[list]
## Next Steps
[if failures]
Rules
- Parallel execution — Launch independent tasks together
- Respect dependencies — Never skip blockedBy
- Report failures — Don't hide issues
$ARGUMENTS