agentleFS
quruhao4800quruhao4800/agent-workflow/skills/workflow/subagent-driven-development/SKILL.md

subagent-driven-development

Skill1 starsChanged 6 months agoMITView on GitHub

Use when executing implementation plans with many tasks or large file impact. Also invoked automatically when executing-plans detects the plan exceeds its threshold (>8 tasks or any task touches >5 files).

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use chinese-code-review, chinese-commit-conventions and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • Leodorareluctant259Skill

    chinese-code-review

    中文代码审查规范——在保持专业严谨的同时,用符合国内团队文化的方式给出有效反馈

    Also in 8 projects that use this

  • Leodorareluctant259Skill

    chinese-commit-conventions

    中文 Git 提交规范 — 适配国内团队的 commit message 规范和 changelog 自动化

    Also in 8 projects that use this

  • obraSkill

    diagnosing-superpowers

    Use when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.

    Also in 19 projects that use this

  • Tommy-ywSkill

    webhook-subscriptions

    Create and manage webhook subscriptions for event-driven agent activation, or for direct push notifications (zero LLM cost). Use when the user wants external services to trigger agent runs OR push notifications to chats.

    Also in 5 projects that use this

  • math-incSkill

    minecraft-modpack-server

    Set up a modded Minecraft server from a CurseForge/Modrinth server pack zip. Covers NeoForge/Forge install, Java version, JVM tuning, firewall, LAN config, backups, and launch scripts.

    Also in 5 projects that use this

  • math-incSkill

    pokemon-player

    Play Pokemon games autonomously via headless emulation. Starts a game server, reads structured game state from RAM, makes strategic decisions, and sends button inputs — all from the terminal.

    Also in 5 projects that use this

  • +3 more from Leodorareluctant259
  • +2 more from math-inc

The file

Show its rules and text

We split this skill into the rules it gives an agent

18 rules. 0 are common, 11 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 179 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If 03-implementation-plan.md already has tasks with status other than pending, a prior session was interrupted.

    line 16
  2. Distinctive:

    Read all task statuses.

    line 18
  3. Distinctive:

    Before dispatching any implementer, scan the full task list for parallelism opportunities:

    line 44
  4. Distinctive:

    When not to parallelize: if you cannot confidently determine the file/table overlap from the plan, default to sequential.

    line 55
  5. Very specific:

    Check Depends on — all listed tasks must be completed before dispatching.

    line 59
  6. Distinctive:

    If questions are raised, answer before implementation.

    line 61
  7. Distinctive:

    If spec issues exist, implementer fixes and spec review repeats.

    line 74
  8. Distinctive:

    If quality issues exist:

    line 76
  9. Very specific:

    Update 04-verification.md evidence for task REQ/CR/DR mapping.

    line 81
  10. Distinctive:

    If a new requirement or optimization appears mid-run:

    line 86
  11. Very specific:

    If scope change: add CR-### in 05-change-log.md, update 01-requirements.md.

    line 89
  12. Very specific:

    If design change only: add DR-### in 05-change-log.md, update 02-design.md.

    line 90
  13. Very specific:

    Update 03-implementation-plan.md tasks/status (pending/in_progress/completed/superseded/cancelled).

    line 91
  14. Very specific:

    Update impacted verification entries in 04-verification.md.

    line 92
  15. Distinctive:

    No task implementation is allowed for unmapped requests.

    line 95
  16. Distinctive:

    Before final reviewer and branch-finishing:

    line 99
  17. Distinctive:

    Start code quality review before spec compliance passes

    line 112
  18. Distinctive:

    Start work on main/master without explicit user consent

    line 115

“Common” is measured: kept by at least 6 in 10 of the 179 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Subagent-Driven Development

Execute one planned task at a time with a fresh implementer subagent, then two reviews: spec compliance first, code quality second.

Core Principle

Fresh subagent per task + strict traceability + two-stage review.

RESUME Mode (when session was interrupted)

If 03-implementation-plan.md already has tasks with status other than pending, a prior session was interrupted. Do not start from scratch.

  1. Read all task statuses.
  2. completed → skip entirely, do not re-implement.
  3. in_progress → check whether the implementer produced any code output (look for commits or file changes since the task was marked in_progress):
    • Code exists → interrupted mid-review. Do not re-implement. Run spec review and quality review on the existing output. If reviews pass, mark completed. If reviews fail, dispatch implementer to fix only the flagged issues.
    • No code exists → interrupted before implementer finished. Treat as pending and dispatch implementer normally.
  4. pending → enter normal dispatch flow.
  5. Announce: "Resuming plan. Completed: N | In-progress (review-only): M | In-progress (re-dispatch): L | Pending: K."

Preflight (MANDATORY)

Step 0: Load project rules: read all files matching rules/**/*.md. Apply them for the entire session.

Load artifacts from one feature folder before dispatching any subagent:

  • 01-requirements.md
  • 02-design.md
  • 03-implementation-plan.md
  • 04-verification.md
  • 05-change-log.md
  • 99-mistake-log.md
  • 06-research-notes.md (if present)

Build a task list from 03-implementation-plan.md and include REQ-###/CR-###/DR-### mapping per task.

Parallel Task Identification (after Preflight)

Before dispatching any implementer, scan the full task list for parallelism opportunities:

  1. Find the ready set: tasks whose Depends on lists are fully completed.
  2. Within the ready set, group tasks that have no shared file targets and no shared DB table modifications.
  3. Tasks in the same group may be dispatched in parallel — send multiple implementer subagents in a single message.
  4. Tasks with shared files or tables must remain sequential even if dependencies are met.

Announce: "Ready set: [Task A, B, C]. Parallel groups: [[A, C], [B] (sequential — shared file XxxMapper.xml)]."

Parallel review: each parallel implementer completes its own spec review and quality review independently. If one implementer's review fails, the others continue uninterrupted. The failed task returns to in_progress for targeted fixes only.

When not to parallelize: if you cannot confidently determine the file/table overlap from the plan, default to sequential. Incorrect parallelism with overlapping writes causes merge conflicts and is harder to recover from than sequential slowness.

Per-Task Flow

  1. Check Depends on — all listed tasks must be completed before dispatching. If not, skip and report blocked.
  2. Dispatch implementer subagent using implementer-prompt.md.
  3. If questions are raised, answer before implementation.
  4. Implementer runs tests and reports outputs.
  5. Implementer verifies every DoD item from 03-implementation-plan.md before handing off:
    • All tests pass; coverage meets targets
    • No TODO / debug / temporary code left
    • Naming follows project rules
    • Error handling complete
    • All callers in Impact handled
    • Commit message: type(scope): one-line English summary

[USER CHECKPOINT 1] Present implementer's concise report to user. Wait for explicit confirmation (e.g. "continue", "ok", "确认") before proceeding to review. Do NOT dispatch reviewers until confirmed.

  1. Dispatch spec reviewer (spec-reviewer-prompt.md).
  2. If spec issues exist, implementer fixes and spec review repeats.
  3. Dispatch code-quality reviewer (code-quality-reviewer-prompt.md).
  4. If quality issues exist:

[USER CHECKPOINT 2] Present review findings (MUST/SHOULD violations) to user. Wait for explicit confirmation before dispatching implementer to fix. User may choose to skip SHOULD fixes.

Implementer fixes confirmed issues and quality review repeats. 10. Update 04-verification.md evidence for task REQ/CR/DR mapping. 11. Mark task complete.

Plan Drift Gate (MANDATORY)

If a new requirement or optimization appears mid-run:

  1. Stop dispatching implementers.
  2. If scope change: add CR-### in 05-change-log.md, update 01-requirements.md. If design change only: add DR-### in 05-change-log.md, update 02-design.md.
  3. Update 03-implementation-plan.md tasks/status (pending/in_progress/completed/superseded/cancelled).
  4. Update impacted verification entries in 04-verification.md.
  5. Re-extract tasks, then resume dispatch.

No task implementation is allowed for unmapped requests.

Completion Gate

Before final reviewer and branch-finishing:

  • unmapped_req = 0
  • open_cr = 0
  • pending_recheck = 0

Then invoke agent-workflow:finishing-a-development-branch.

Red Flags

Never:

  • Skip either review stage
  • Start code quality review before spec compliance passes
  • Move to next task with unresolved reviewer issues
  • Implement requirement changes without CR + plan updates
  • Start work on main/master without explicit user consent

Integration

  • REQUIRED: agent-workflow:writing-plans
  • REQUIRED: agent-workflow:requesting-code-review
  • REQUIRED: agent-workflow:finishing-a-development-branch
  • Subagents should use: agent-workflow:test-driven-development

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.