subagent-driven-development

---

name: subagent-driven-development description: Use when executing implementation plans with independent tasks in the current session. Triggers: multi-task impl plan, independent tasks, same session, parallel implementation, 子代理开发, subagent development.

Subagent-Driven Development

Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.

Core principle: Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration

When to Use

digraph when_to_use {
    "Have implementation plan?" [shape=diamond];
    "Tasks mostly independent?" [shape=diamond];
    "Stay in this session?" [shape=diamond];
    "subagent-driven-development" [shape=box];
    "executing-plans" [shape=box];
    "Manual execution or brainstorm first" [shape=box];

    "Have implementation plan?" -> "Tasks mostly independent?" [label="yes"];
    "Have implementation plan?" -> "Manual execution or brainstorm first" [label="no"];
    "Tasks mostly independent?" -> "Stay in this session?" [label="yes"];
    "Tasks mostly independent?" -> "Manual execution or brainstorm first" [label="no - tightly coupled"];
    "Stay in this session?" -> "subagent-driven-development" [label="yes"];
    "Stay in this session?" -> "executing-plans" [label="no - parallel session"];
}

Workflow Detection

If docs/tasks/<task-id>.yaml exists and status == execute, this skill is running within /run-ce-workflow. In that case, the orchestrator manages state transitions — skip the final handoff to finishing-a-development-branch and instead set stages.execute.all_checkpoints_done: true and stages.execute.all_tests_green: true on completion.

Workflow

1. Read the implementation plan and identify independent tasks.
2. Create a TodoWrite checklist with one item per task (and any session setup items).
3. For each task: dispatch an implementer subagent (after optional AskQuestion confirmation), then run spec review and code quality review loops until approved.
4. Review each subagent’s output; resolve NEEDS_CONTEXT, BLOCKED, or DONE_WITH_CONCERNS before moving on.
5. Integrate results across tasks, run final review, and verify (e.g. tests, workflow completion skills).

Create TodoWrite todos from this workflow (use task-scoped IDs, e.g. sadd-{task}-1):

TodoWrite todos:
  - id: "sadd-setup-1", content: "Read plan + list independent tasks", status: "pending"
  - id: "sadd-setup-2", content: "TodoWrite checklist created", status: "pending"
  - id: "sadd-task-1", content: "Task 1: dispatch → reviews → complete", status: "pending"
  - id: "sadd-final-1", content: "Final review + integrate + verify", status: "pending"

The Process

digraph process {
    rankdir=TB;

    subgraph cluster_per_task {
        label="Per Task";
        "Dispatch implementer" [shape=box];
        "Questions?" [shape=diamond];
        "Answer, re-dispatch" [shape=box];
        "Implement + test + commit" [shape=box];
        "Spec review" [shape=box];
        "Spec OK?" [shape=diamond];
        "Fix spec gaps" [shape=box];
        "Code quality review" [shape=box];
        "Quality OK?" [shape=diamond];
        "Fix quality issues" [shape=box];
        "Mark complete" [shape=box];
    }

    "Read plan, create TodoWrite" [shape=box];
    "More tasks?" [shape=diamond];
    "Final review" [shape=box];
    "superpowers:finishing-a-development-branch" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, create TodoWrite" -> "Dispatch implementer";
    "Dispatch implementer" -> "Questions?";
    "Questions?" -> "Answer, re-dispatch" [label="yes"];
    "Answer, re-dispatch" -> "Dispatch implementer";
    "Questions?" -> "Implement + test + commit" [label="no"];
    "Implement + test + commit" -> "Spec review";
    "Spec review" -> "Spec OK?";
    "Spec OK?" -> "Fix spec gaps" [label="no"];
    "Fix spec gaps" -> "Spec review" [label="re-review"];
    "Spec OK?" -> "Code quality review" [label="yes"];
    "Code quality review" -> "Quality OK?";
    "Quality OK?" -> "Fix quality issues" [label="no"];
    "Fix quality issues" -> "Code quality review" [label="re-review"];
    "Quality OK?" -> "Mark complete" [label="yes"];
    "Mark complete" -> "More tasks?";
    "More tasks?" -> "Dispatch implementer" [label="yes"];
    "More tasks?" -> "Final review" [label="no"];
    "Final review" -> "superpowers:finishing-a-development-branch";
}

Task Start Confirmation

Before dispatching each implementation subagent, use the AskQuestion tool:

• title: "Task {N}: {task name}"
• question prompt: "Ready to dispatch implementation subagent for:\n{brief task description}"
• options:
  • "Proceed with this task"
  • "Skip this task"
  • "Reorder — do a different task next"
  • "Approve all remaining tasks without pausing"

If user selects "Approve all remaining tasks", skip AskQuestion for subsequent tasks. Still show task summaries before dispatching.

Model Selection

Use the least powerful model that can handle each role.

Task type	Cursor model param	Signals
Mechanical implementation (1-2 files, clear spec)	fast	Isolated function, well-specified
Integration / judgment (multi-file, debugging)	(default)	Pattern matching, cross-file coordination
Architecture / design / review	(default)	Broad codebase understanding needed

Handling Implementer Status

Implementer subagents report one of four statuses. Handle each appropriately:

DONE: Proceed to spec compliance review.

DONE_WITH_CONCERNS: The implementer completed the work but flagged doubts. Read the concerns before proceeding. If the concerns are about correctness or scope, address them before review. If they're observations (e.g., "this file is getting large"), note them and proceed to review.

NEEDS_CONTEXT: The implementer needs information that wasn't provided. Provide the missing context and re-dispatch.

BLOCKED: The implementer cannot complete the task. Assess the blocker:

1. If it's a context problem, provide more context and re-dispatch with the same model
2. If the task requires more reasoning, re-dispatch with a more capable model
3. If the task is too large, break it into smaller pieces
4. If the plan itself is wrong, escalate to the human

Never ignore an escalation or force the same model to retry without changes. If the implementer said it's stuck, something needs to change.

Prompt Templates

Read the prompt template files from this skill's directory and pass their content as the Task tool's prompt parameter:

• ./implementer-prompt.md — Dispatch implementer subagent
• ./spec-reviewer-prompt.md — Dispatch spec compliance reviewer subagent
• ./code-quality-reviewer-prompt.md — Dispatch code quality reviewer subagent

Example Workflow

Step	Action	Result
Setup	Read plan, extract all tasks, create TodoWrite	5 tasks queued
Task 1	Dispatch implementer → asks question → answer → implements	5/5 tests passing, committed
	Dispatch spec reviewer	✅ Spec compliant
	Dispatch code quality reviewer	✅ Approved
Task 2	Dispatch implementer → implements	8/8 tests passing, committed
	Dispatch spec reviewer	❌ Missing progress reporting, extra --json flag
	Implementer fixes → spec reviewer re-reviews	✅ Spec compliant
	Dispatch code quality reviewer	❌ Magic number
	Implementer fixes → code quality re-reviews	✅ Approved
...	Repeat for remaining tasks
Final	Dispatch final code reviewer for entire implementation	Ready to merge

Red Flags

Never:

• Start implementation on main/master branch without explicit user consent
• Skip reviews (spec compliance OR code quality)
• Proceed with unfixed issues
• Dispatch multiple implementation subagents in parallel (conflicts)
• Make subagent read plan file (provide full text instead)
• Skip scene-setting context (subagent needs to understand where task fits)
• Ignore subagent questions (answer before letting them proceed)
• Accept "close enough" on spec compliance (spec reviewer found issues = not done)
• Skip review loops (reviewer found issues = implementer fixes = review again)
• Let implementer self-review replace actual review (both are needed)
• Start code quality review before spec compliance is ✅ (wrong order)
• Move to next task while either review has open issues
• Forgetting to offer "Approve all remaining" — don't force user to click through every task if they trust the plan

If subagent asks questions:

• Answer clearly and completely
• Provide additional context if needed
• Don't rush them into implementation

If reviewer finds issues:

• Implementer (same subagent) fixes them
• Reviewer reviews again
• Repeat until approved
• Don't skip the re-review

If subagent fails task:

• Dispatch fix subagent with specific instructions
• Don't try to fix manually (context pollution)

Integration

Required workflow skills:

• superpowers:using-git-worktrees — Set up isolated workspace before starting
• superpowers:writing-arch-plans / superpowers:writing-impl-plans — Creates the plan this skill executes
• superpowers:requesting-code-review — Code review template for reviewer subagents
• superpowers:finishing-a-development-branch — Complete development after all tasks

Subagents should use:

• superpowers:test-driven-development — Subagents follow TDD for each task

Alternative workflow:

• superpowers:executing-plans — Use for parallel session instead of same-session execution