← 返回 Skills 市场
1136
总下载
2
收藏
2
当前安装
1
版本数
在 OpenClaw 中安装
/install delegate
功能描述
Route tasks to sub-agents with optimal model selection, error recovery, and result verification.
使用说明 (SKILL.md)
Core Rule
Spawn cost \x3C task cost → delegate. Otherwise, do it yourself.
Model Tiers
| Tier | Models | Cost | Use for |
|---|---|---|---|
| Small | Haiku, GPT-4o-mini, Gemini Flash | ~$0.25/1M | Search, summarize, format, classify |
| Medium | Sonnet, GPT-4o, Gemini Pro | ~$3/1M | Code, analysis, synthesis |
| Large | Opus, o1, Gemini Ultra | ~$15/1M | Architecture, complex reasoning |
Rule of thumb: Start with smallest tier. Escalate only if output quality insufficient.
Spawn Checklist
Every spawn must include:
1. TASK: Single clear deliverable (not "help with X")
2. MODEL: Explicit tier choice
3. CONTEXT: Only files/info needed (never full history)
4. OUTPUT: Expected format ("return JSON with...", "write to file X")
5. DONE: How to signal completion
Check templates.md for copy-paste spawn templates.
Error Recovery
| Error Type | Action |
|---|---|
| Sub-agent timeout (>5 min no response) | Kill and retry once |
| Wrong output format | Retry with stricter instructions |
| Task too complex for tier | Escalate: Small→Medium→Large |
| Repeated failures (3x) | Abort, report to user |
Check errors.md for recovery patterns and escalation logic.
Verification
Never trust "done" without checking:
- Code: Run tests, check syntax
- Files: Verify they exist and have content
- Data: Spot-check 2-3 items
- Research: Confirm sources exist
Don't Delegate
- Quick tasks (\x3C30 seconds to do yourself)
- Tasks needing conversation context
- Anything requiring user clarification mid-task
安全使用建议
This skill is an instruction-only delegator that will tell the agent how and when to spawn sub-agents and what context to give them. Before installing or enabling it: 1) Ask the author for source/homepage or a formal security/privacy statement. 2) Require explicit user confirmation before any spawned sub-agent is given access to local files, environment variables, repo write/commit permissions, or credentials. 3) Test in a restricted/isolated workspace (no secrets) to observe behavior. 4) Add policies or runtime guards that block access to secrets or global config unless the user explicitly approves which specific files/vars may be shared. 5) If you must use it, prefer manual invocation and monitor logs for file writes or outbound requests. These precautions reduce the risk that delegated tasks will accidentally or intentionally exfiltrate sensitive data.
功能分析
Type: OpenClaw Skill
Name: delegate
Version: 1.0.0
The OpenClaw AgentSkills skill bundle 'delegate' is designed to manage and route tasks to sub-agents. While this involves powerful capabilities such as code execution, file system access, and network operations (as seen in `templates.md`), the skill itself does not exhibit malicious intent. Crucially, `SKILL.md` includes security-conscious instructions like 'CONTEXT: Only files/info needed (never full history)' to limit information exposure to sub-agents, and 'Verification' steps to prevent blind trust of sub-agent outputs. The instructions are clear, focused on efficient task delegation and error recovery (`errors.md`), and lack any evidence of data exfiltration, persistence, unauthorized actions, or prompt injection attempts against the main agent.
能力评估
Purpose & Capability
The name/description match the instructions: templates and checklists are designed to spawn sub-agents, select model tiers, escalate on failures, and verify outputs — all coherent with a delegator skill. However, the skill has no source/homepage and declares no required env/config access even though the templates and error patterns explicitly reference including file contexts, env vars, and paths when troubleshooting.
Instruction Scope
SKILL.md and templates repeatedly instruct the agent to include 'only files/info needed' but also, in errors.md and templates.md, recommend including env vars and paths to fix failures (e.g., 'Include env vars, paths', 'Read [relevant files]', 'Commit with message...'). That gives the skill (and any spawned sub-agent) broad discretion to read local files, environment variables, and modify/commit files — behavior beyond a simple delegation interface and not explicitly constrained.
Install Mechanism
Instruction-only skill with no install spec and no code files; nothing is written to disk by an installer. This is low-risk from an install-mechanism perspective.
Credentials
The registry metadata declares no required environment variables or credentials, yet errors.md and templates indicate that env vars and paths may be included as needed. That mismatch means the skill could legitimately request access to secrets or config at runtime without having declared them up-front, which is disproportionate and risky.
Persistence & Privilege
always:false and no install steps; the skill does not request persistent presence. It can invoke sub-agents autonomously (platform default) — combined with the instruction-scope concerns this increases potential blast radius, but autonomous invocation alone is expected.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install delegate - 安装完成后，直接呼叫该 Skill 的名称或使用
/delegate触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.0
Initial release - sub-agent routing with error recovery
元数据
常见问题
Delegate 是什么？
Route tasks to sub-agents with optimal model selection, error recovery, and result verification. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 1136 次。
如何安装 Delegate？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install delegate」即可一键安装，无需额外配置。
Delegate 是免费的吗？
是的，Delegate 完全免费（开源免费），可自由下载、安装和使用。
Delegate 支持哪些平台？
Delegate 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Delegate？
由 Iván（@ivangdavila）开发并维护，当前版本 v1.0.0。
推荐 Skills