Borrowing it
Nothing to install: this file belongs to leenkcool/humanasllm. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/leenkcool/humanasllm/main/.claude/skills/dispatch-human/SKILL.mdgit clone --depth 1 https://github.com/leenkcool/humanasllmWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/leenkcool/humanasllm/dispatch-human)<a href="https://agentmods.dev/skills/leenkcool/humanasllm/dispatch-human"><img src="https://agentmods.dev/badge/skills/leenkcool/humanasllm/dispatch-human/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/leenkcool/humanasllm/dispatch-human"><img src="https://agentmods.dev/badge/skills/leenkcool/humanasllm/dispatch-human.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00083 | $0.01304 |
| Opus 5.5 | $0.00033 | $0.00522 |
| Sonnet 5.5 | $0.00017 | $0.00261 |
| Haiku 4.5 | $0.00008 | $0.00130 |
Grade A, and why
dispatch-human scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
派单给人工工程师(dispatch-human)
本 Skill 是「humanllm agent」的触发入口:它自己不实现任务,而是整理触发条件 → 委派给 humanllm agent → 回传人工产出。
Trigger(何时触发)
- 用户明确要求人工处理:「交给人工」「派给人类」「让工程师做」「需要真人实现」
- 任务涉密 / 私有内部逻辑 / 不可外传 / 敏感业务
- 项目规则(CLAUDE.md)明确要求人工编写
- 任务可能额外需要外部资源(服务器 / API Key / 权限 / 环境)——humanllm 会自动先提审批
Steps(主代理执行)
- 读取 调用者任务原文与全部上下文(messages、约束、涉密标记)。
- 标注交人工原因:
安全原因或项目规则要求,随任务一并传给 agent。 - 委派:调用 Agent 工具,参数如下:
humanllm 会按subagent_type: "humanllm" prompt: <任务原文 + 交人工原因标注> run_in_background: false // /v1 异步受理,humanllm 登记未完成、之后回查交付;大任务可后台.claude/agents/humanllm.md自动把任务整理成「完整上下文包」(【任务】【交人工原因】【项目与代码库】【接单流程】【环境约定】【要求】),写入data/human_task.json并提交 p390 网关。 - 登记未完成(humanllm 执行):/v1 异步受理立即返回
task_id(人工小时级,不再挂起等待)。humanllm 把 task_id 登记到data/human_followup.json(pending)。 - 回查交付(humanllm 执行):后续做相关工作再次调用时,humanllm 先回查
GET /v1/tasks/:id——completed把人工产出逐字返回(保留代码块与换行);未完成则如实反馈状态、保留登记继续轮候。
主代理只负责「判断触发 → 委派 → 接收产出/状态」,不直接读写
data/下任何文件;登记、回查、交付全部由 humanllm 按.claude/agents/humanllm.md自动完成。
跟进时机(何时再次委派 humanllm 回查)
humanllm 每次被委派都会先回查 data/human_followup.json 里的未完成任务。因此:
- 用户再次提问、做与之相关的工作、或派新任务时,主代理先委派一次 humanllm(prompt 注明「先回查未完成的人工任务,再处理本次请求」),humanllm 会先交付已完成的产出、反馈未完成的状态,再继续本次工作。
- 只要对话仍在进行,就把「跟进人工任务」作为相关工作开始前的固定动作,防遗忘。
委派参数速查
| 参数 | 值 | 说明 |
|---|---|---|
subagent_type |
humanllm |
指定人工代理 agent |
run_in_background |
false |
异步受理:humanllm 登记未完成、跟进时回查交付;大任务可后台 true |
model |
省略 | humanllm 定义已定(转发器,仅 haiku 占位) |
isolation |
省略 | 无需 worktree(转发类任务) |
Verification(怎么算做对)
- ✅ 确实调用了
humanllmagent(而非自行实现任务)。 - ✅ 任务包六要素齐全(humanllm 保证):【任务】【交人工原因】【项目与代码库】【接单流程】【环境约定】【要求】。
- ✅ 拿到
task_id已登记到data/human_followup.json(未完成,防遗忘)。 - ✅ 回传内容是人工产出原文(
completed回查),未自行改写、未丢代码块/换行。 - ✅ 未完成任务如实反馈状态(pending/processing/returned),未谎称完成,继续轮候。
- ✅ 附上了任务编号(如 #21)与工作台地址
http://192.168.168.3:39000/login.html。
硬绑定变体(可选)
若希望整个 skill 直接以 humanllm agent 身份运行(跳过"主代理→委派"),把 frontmatter 改为:
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2mo ago First seen · 75 lines · 83 tokens per session scan A f6e57cd8d266
dispatch-human is a skill published in the GitHub repository leenkcool/humanasllm (29 stars, last pushed 7d ago), licensed MIT. It adds 83 tokens to every session and 1,304 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
dispatch-human compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| dispatch-humanleenkcool/humanasllm | Skill | 29 | 7d ago | MIT | A | 83 | Claude Code |
| browser-usebrowser-use/browser-use | Skill | 120k | yesterday | MIT | A | 57 | Claude Code, Codex |
| open-sourcebrowser-use/browser-use | Skill | 120k | yesterday | MIT | A | 137 | Claude Code |
| crit-storytomasz-tomczyk/crit | Skill | 1.3k | 2d ago | MIT | A | 47 | Claude Code, Codex |
| snipedouard-claude/snip | Skill | 478 | yesterday | MIT | A | 0 | Claude Code, Codex |
| goaizendev-sh/goai | Skill | 234 | 6d ago | MIT | A | 0 | Claude Code, Codex |
| generate-rfcsnurettincoban/ai-prd-workflow | Skill | 309 | 29d ago | MIT | A | 34 | Claude Code |