subagent-driven-development

name	subagent-driven-development
description	Use when executing implementation plans with independent tasks in the current session

Delegate-Driven Development

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

Why delegates: You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.

Core principle: Fresh delegate 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];
    "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?" -> "subagent-driven-development" [label="yes"];
    "Tasks mostly independent?" -> "Manual execution or brainstorm first" [label="no - tightly coupled"];
}

The Process

digraph process {
    rankdir=TB;

    subgraph cluster_per_task {
        label="Per Task";
        "Dispatch implementer delegate" [shape=box];
        "Implementer asks questions?" [shape=diamond];
        "Answer questions, provide context" [shape=box];
        "Implementer implements, tests, self-reviews" [shape=box];
        "Dispatch spec reviewer delegate" [shape=box];
        "Spec reviewer confirms code matches spec?" [shape=diamond];
        "Implementer fixes spec gaps" [shape=box];
        "Dispatch code quality reviewer delegate" [shape=box];
        "Code quality reviewer approves?" [shape=diamond];
        "Implementer fixes quality issues" [shape=box];
        "Mark task complete" [shape=box];
    }

    "Read plan, extract all tasks with full text, note context" [shape=box];
    "More tasks remain?" [shape=diamond];
    "Dispatch final code reviewer delegate for entire implementation" [shape=box];
    "Implementation complete" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, extract all tasks with full text, note context" -> "Dispatch implementer delegate";
    "Dispatch implementer delegate" -> "Implementer asks questions?";
    "Implementer asks questions?" -> "Answer questions, provide context" [label="yes"];
    "Answer questions, provide context" -> "Dispatch implementer delegate";
    "Implementer asks questions?" -> "Implementer implements, tests, self-reviews" [label="no"];
    "Implementer implements, tests, self-reviews" -> "Dispatch spec reviewer delegate";
    "Dispatch spec reviewer delegate" -> "Spec reviewer confirms code matches spec?";
    "Spec reviewer confirms code matches spec?" -> "Implementer fixes spec gaps" [label="no"];
    "Implementer fixes spec gaps" -> "Dispatch spec reviewer delegate" [label="re-review"];
    "Spec reviewer confirms code matches spec?" -> "Dispatch code quality reviewer delegate" [label="yes"];
    "Dispatch code quality reviewer delegate" -> "Code quality reviewer approves?";
    "Code quality reviewer approves?" -> "Implementer fixes quality issues" [label="no"];
    "Implementer fixes quality issues" -> "Dispatch code quality reviewer delegate" [label="re-review"];
    "Code quality reviewer approves?" -> "Mark task complete" [label="yes"];
    "Mark task complete" -> "More tasks remain?";
    "More tasks remain?" -> "Dispatch implementer delegate" [label="yes"];
    "More tasks remain?" -> "Dispatch final code reviewer delegate for entire implementation" [label="no"];
    "Dispatch final code reviewer delegate for entire implementation" -> "Implementation complete";
}

Model Selection

Use the least powerful model that can handle each role to conserve cost and increase speed.

Mechanical implementation tasks (isolated functions, clear specs, 1-2 files): use a fast, cheap model. Most implementation tasks are mechanical when the plan is well-specified.

Integration and judgment tasks (multi-file coordination, pattern matching, debugging): use a standard model.

Architecture, design, and review tasks: use the most capable available model.

Task complexity signals:

• Touches 1-2 files with a complete spec → cheap model
• Touches multiple files with integration concerns → standard model
• Requires design judgment or broad codebase understanding → most capable model

Handling Implementer Status

Implementer delegates 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.

Example Workflow

You: I'm using Delegate-Driven Development to execute this plan.

[Read plan file once: docs/specs/<feature>/plan.md]
[Extract all 5 tasks with full text and context]

Task 1: Hook installation script

[Dispatch implementer delegate with full task text + context]

Implementer: "Before I begin - should the hook be installed at user or system level?"

You: "User level (~/.config/hooks/)"

Implementer: "Got it. Implementing now..."
[Later] Implementer:
  - Implemented install-hook command
  - Added tests, 5/5 passing
  - Self-review: Found I missed --force flag, added it
  - Reported DONE

[Dispatch spec compliance reviewer delegate]
Spec reviewer: Spec compliant - all requirements met, nothing extra

[Get git SHAs, dispatch code quality reviewer delegate]
Code reviewer: Strengths: Good test coverage, clean. Issues: None. Approved.

[Mark Task 1 complete]

Task 2: Recovery modes

[Dispatch implementer delegate with full task text + context]

Implementer: [No questions, proceeds]
Implementer:
  - Added verify/repair modes
  - 8/8 tests passing
  - Self-review: All good
  - Reported DONE

[Dispatch spec compliance reviewer delegate]
Spec reviewer: Issues:
  - Missing: Progress reporting (spec says "report every 100 items")
  - Extra: Added --json flag (not requested)

[Implementer fixes issues]
Implementer: Removed --json flag, added progress reporting

[Spec reviewer reviews again]
Spec reviewer: Spec compliant now

[Dispatch code quality reviewer delegate]
Code reviewer: Strengths: Solid. Issues (Important): Magic number (100)

[Implementer fixes]
Implementer: Extracted PROGRESS_INTERVAL constant

[Code reviewer reviews again]
Code reviewer: Approved

[Mark Task 2 complete]

...

[After all tasks]
[Dispatch final devops-reviewer delegate]
Final reviewer: All requirements met, ready to merge

Done!

Advantages

vs. Manual execution:

• Delegates follow TDD naturally
• Fresh context per task (no confusion)
• Parallel-safe (delegates don't interfere)
• Delegate can ask questions (before AND during work)

Efficiency gains:

• No file reading overhead (controller provides full text)
• Controller curates exactly what context is needed
• Delegate gets complete information upfront
• Questions surfaced before work begins (not after)

Quality gates:

• Self-review catches issues before handoff
• Two-stage review: spec compliance, then code quality
• Review loops ensure fixes actually work
• Spec compliance prevents over/under-building
• Code quality ensures implementation is well-built

Cost:

• More delegate invocations (implementer + 2 reviewers per task)
• Controller does more prep work (extracting all tasks upfront)
• Review loops add iterations
• But catches issues early (cheaper than debugging later)

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 delegates in parallel on SHARED files (conflicts). Parallel dispatch is fine when tasks touch distinct files — use the execution-planning skill to verify independence before parallelizing
• Make delegate read plan file (provide full text instead)
• Skip scene-setting context (delegate needs to understand where task fits)
• Ignore delegate 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 passes (wrong order)
• Move to next task while either review has open issues
• Let subagents run git commit or git push (orchestrator handles all git operations)

If delegate asks questions:

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

If reviewer finds issues:

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

If delegate fails task:

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

Task Enrichment (before dispatch)

Before dispatching each task, enrich the task description with context accumulated from earlier tasks:

1. Re-read the task description from the plan file (don't rely on memory)
2. Add context discovered from earlier tasks: new file paths created, API signatures found, config values, gotchas encountered
3. If the task description is too vague for a subagent to succeed without judgment calls, update the plan file with the enriched description BEFORE dispatching
4. The plan file should always reflect what the subagent actually receives — not the original sparse description

The goal: a subagent reading only the enriched task description should be able to execute it mechanically.

Document Tracking (after each task)

After each task completes (reaches "Mark task complete"):

1. Update the plan file: change - [ ] → - [x] for the completed task
2. If the task was implemented differently than planned, append an Actual: note to the task entry explaining what actually happened
3. If the task description was enriched before dispatch, ensure the plan file reflects the enriched version (not the original sparse text)
4. If new tasks were added mid-implementation, add them to the plan file before continuing
5. If the approach changed fundamentally, update the plan's architecture section

The plan file is the single source of truth for "what happened." Someone reading it after implementation should understand both the intent and the reality.

Related Skills

• writing-plans — Creates the plan this skill executes
• test-driven-development — Delegates should follow TDD for each task
• verification-before-completion — Verify before claiming work is done