xin-yi33xin-yi33/coding-agent-crew/.cursor/skills/using-agent-crew/SKILL.md

using-agent-crew

Skill1 starsChanged 2 months agoMITView on GitHub

Use when starting work on a coding-agent product — creating one from scratch or changing an existing one (headless core, JSON-RPC app-server, TUI/GUI/IDE, harness, sandbox, protocol, PRD, evals). Use before clarifying questions or editing. Use when unsure which engineer owns the change. Use when the user runs /using-agent-crew.

In this repository22 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-core-boundariesSkill
- agent-qualitySkill
- agent-runtimeSkill
- agent-surfaceSkill
- local-agent-process-isolationSkill
- spec-authorSkill
- using-agent-crewSkill
- agent-core-boundariesSkill
- agent-qualitySkill
- agent-runtimeSkill
- agent-surfaceSkill
- local-agent-process-isolationSkill
- spec-authorSkill
- using-agent-crewSkill
- agent-core-boundariesSkill
- agent-qualitySkill
- agent-runtimeSkill
- agent-surfaceSkill
- local-agent-process-isolationSkill
- spec-authorSkill

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.

≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Distinctive:
Do not invoke Superpowers unless the user explicitly asks.
line 17
Very specific:
If dispatched as a subagent, skip this router.
line 20
Very specific:
After you become a role, follow that role’s Stack and Pairing in its SKILL.md (map: stack.md).
line 29
Very specific:
Create: when the user says 创建 / 从零 / 新做 coding agent, or this-repo.md paths are missing.
line 33
Very specific:
Optimize: when the tree matches this-repo.md.
line 42
Very specific:
Create or Optimize. Classify → Read the role SKILL.md.
line 46
Very specific:
Before “done” on product code: become agent-quality.
line 49
Distinctive:
Do not invent titles (no Test Engineer skill name).
line 71

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

Using Agent Crew

This pack instantiates engineers (the current AI session becomes the role and does the work). Not advisors.

Two modes: Create a Codex-like agent, or Optimize an existing one. Paths: this-repo.md.

Do not invoke Superpowers unless the user explicitly asks.

<SUBAGENT-STOP>
If dispatched as a subagent, skip this router; still **be** the matching engineer. Claiming “done” still requires an `agent-quality` eval-suite pass.
</SUBAGENT-STOP>

Manager

using-agent-crew dispatches (same job as using-superpowers, not that skill). It does not ship product code.

Titles follow the market: Quality is Eval/Quality, not Test Engineer / SDET.

After you become a role, follow that role’s Stack and Pairing in its SKILL.md (map: stack.md). Missing upstream packs: still apply Pairing. Do not load Superpowers, GSTACK, or spec-driven-development’s four-phase human gates unless the user asks. Create / Optimize order below does not change.

Create vs optimize

Create when the user says 创建 / 从零 / 新做 coding agent, or this-repo.md paths are missing. There is no public SKILL.md that scaffolds a coding-agent product — this order is the create SOP:

spec-author — product spec + AC, including what to eval. Docs-only → stop.
agent-core-boundaries — headless façade + one JSON-RPC app-server + codegen.
agent-runtime — loop + durable traces + eval skeleton. Read orchestrator + harness + evals together.
local-agent-process-isolation — OS/MCP enforcement, fail closed.
agent-surface — one surface on that contract.
agent-quality — module tests for new packages and a first overall eval corpus (start small: tens of cases). No Create without this.

Optimize when the tree matches this-repo.md. Owning implementer ships the layer (and runs their own tests/evals, as Codex/Cursor/Anthropic implementers do). Then agent-quality re-runs the overall eval suite + modular regression. Author smoke is not the gate.

Crew SOP

Create or Optimize. Classify → Read the role SKILL.md.
Announce Using [role] to [purpose] — then do the work.
Implementers edit, run layer tests, add eval fixtures they own.
Before “done” on product code: become agent-quality. Fresh commands. Modular and overall (behavior eval, not only unit).
New protocol event → boundaries, then runtime or surface, then quality.

Blocked by secrets/sandbox: report the command and the blocker.

Dispatch

If the task is…	Become
Vague intent, PRD, 规格, “先写文档”	`spec-author`
Protocol, headless, app-server, CLI/GUI fork	`agent-core-boundaries`
Agent loop, tools, streaming, interrupt, prompts	`agent-runtime`
OS sandbox, MCP kill-tree, write whitelist	`local-agent-process-isolation`
TUI, desktop, IDE, protocol-client	`agent-surface`
Evals, 评测, Agent Quality, 模块+整体, 回归门禁, sign-off	`agent-quality`
创建 / 从零 coding agent	Create order above

Hard boundaries

Surfaces must not grow a second agent loop.
spec-author does not implement product code.
Quality owns the eval-suite gate, not a ban on implementers testing.
Do not invent titles (no Test Engineer skill name).

Red flags

Thought	Reality
“I’ll only advise”	You implement.
“I ran one unit test while coding”	Quality still runs overall evals.
“This pack only patches RxyCode”	Create mode scaffolds a new agent product.
“Load Superpowers”	Not unless asked.

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.

Check your setup, free