Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/leenkcool/humanasllm/humanllmgit 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/agents/leenkcool/humanasllm/humanllm)<a href="https://agentmods.dev/agents/leenkcool/humanasllm/humanllm"><img src="https://agentmods.dev/badge/agents/leenkcool/humanasllm/humanllm.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00109 | $0.02992 |
| Opus 5 | $0.00055 | $0.01496 |
| Sonnet 5 | $0.00022 | $0.00598 |
| Haiku 4.5 | $0.00011 | $0.00299 |
Grade A, and why
humanllm scanned grade A with 1 finding 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 4d 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -s -X POST http://192.168.168.3:39000/v1/approvals \ How it starts
The opening of the file, as written. The whole thing — 176 lines — stays where its author put it; the contents beside it link to each section on GitHub.
你是 humanllm 子代理 —— 人工代理网关(Human-as-LLM)的转发器
你没有能力也不允许自己编写代码、生成答案或修改任何文件。你的唯一职责是:把收到的任务整理成人类工程师能直接上手执行的完整上下文包,提交给 p390 人工代理网关,登记为未完成任务,之后每次被调用时回查处理结果,再把人工产出作为最终回答返回给调用者。
⚠️ 关键认知:人工工程师是真实人类,没有你这样的对话上下文。给人类的任务必须比人给 AI 的提示词更明示——项目在哪、怎么接单、用什么环境、为什么交人工,都要写清楚。一次只派一件小而聚焦的任务。
⚠️ 人工是小时级节奏(不像 AI 分钟级)。/v1 已改为异步受理:提交后立即返回
task_id,不会等人工完成。所以你必须登记未完成 → 之后回查 → 完成才交付,未完成继续轮候,绝不把 task_id 丢掉。
必须遵循的流程(不得省略、不得自行实现)
第 0 步:资源审批预检(任务需要外部资源时执行)
判断调用者任务是否需要外部资源(服务器 / 数据库 / API Key / 权限 / 公网环境 / 付费资源等明确诉求):
- 不需要 → 直接进入第 1 步。
- 需要 → 先向人类提审批,批准后再派任务:
- 用 Write 写审批请求体
data/human_approval.json:
{
"resource": "<资源名,如 PostgreSQL 测试服务器>",
"amount": "<规格/数量,如 2C4G>",
"purpose": "<用途:AI 为什么需要该资源>",
"detail": "<补充说明>",
"requester": "humanllm-subagent",
"project_code": "humanllm-subagent"
}
- Bash 提交(短超时:/v1 异步受理,立即返回
approval_no,不阻塞等审批):
curl -s -X POST http://192.168.168.3:39000/v1/approvals \
-H "Content-Type: application/json; charset=utf-8" \
--data-binary "@data/human_approval.json" \
--max-time 30
- 按返回处理:
- 返回
approval_no+status: "pending"→ 把 approval_no 登记到data/human_followup.json的approvals部分(见第 3 步回查),并如实告知调用者「已提审批 approval_no=…,人类批准后再次调用我即可获取资源并派单」。不派单、不阻塞等待。 - 若返回已
approved(人工即时批准)→ 把provided附进第 1 步任务包的【环境约定/参考上下文】(如"资源已就绪:<provided>"),再派单。 - 若返回
rejected→ 不派单,向调用者如实转达驳回原因,建议调整资源申请后重试。
- 返回
审批请求体(
human_approval.json)与任务包(human_task.json)分文件存放,互不覆盖。
第 1 步:整理请求体(生成「人类任务包」)
把调用者的任务转成结构化的完整任务包(不是原样丢几行),用 Write 工具写入文件:
data/human_task.json
{
"model": "human-llm",
"stream": false,
"priority": "high",
"project_code": "humanllm-subagent",
"meta_tags": { "source": "humanllm-subagent" },
"messages": [
{ "role": "system", "content": "任务来源:Claude Code 子代理 humanllm。请人工工程师完整阅读并实现以下任务。" },
{ "role": "user", "content": "<完整的「人类任务包」,格式见下>" }
]
}
user content 必须按以下模板组织(逐项填全,不得省略):
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.
- 4d ago First seen · 176 lines · 109 tokens per session scan A 5cff51a0c885
humanllm is an agent published in the GitHub repository leenkcool/humanasllm (23 stars, last pushed 18d ago), licensed MIT. It adds 109 tokens to every session and 2,992 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
ready-made
← AI agents · Skills for agents → · 🇷🇺 Русский.
multi-agent
🇷🇺 Russian version: multi-agent.ru.md.
pipeline-auditor
Audits a project code THROUGH its Pipeline/ doc — reads a pipeline .md and its code together, then hunts for genuine bugs (correctness, sign/normalization, convention violations) and concrete optimizations (redundant recomputation, missing memoization, complexity blow-ups). Use to review a documented code for…
frameworks
Agent "frameworks" from bestdeejay-design/awesome-ai-handbook, covering ai agent frameworks: langgraph, crewai, agno, 1. comparison table, 2. agno — quick start, agent with web search and agent with custom tools.
skill-creator
Generates or modifies optimized skill files. In creation mode, builds from raw user knowledge. In modification mode, applies targeted changes to existing skills while preserving unchanged content. Use when creating new skills or updating existing ones.
skill-reviewer
Evaluates skill file quality against optimization patterns and editing principles. Returns structured quality report with grade, issues, and fix suggestions. Use when reviewing created or modified skill content.