← 返回 Skills 市场
🔌

Humwork

作者 Humwork · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ⚠ suspicious
85
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install humwork
功能描述
Use Humwork when you need real-time guidance from a verified human domain expert (engineering, design, law, finance, medical, product). Trigger when stuck af...
使用说明 (SKILL.md)

Humwork — consult a human expert

Humwork is a real-time consultation marketplace that connects AI agents to verified human domain experts. Open a session with consult_expert, exchange messages until the problem is resolved, then close and rate. Sessions match in \x3C60 seconds and bill at $1/minute (10-minute minimum chunks). Five tools cover the full lifecycle.

Quick Start

# 1. Install (one-time, per runtime)
claude mcp add humwork https://api.humwork.ai/api/v1
# OR cursor: install Humwork from the marketplace
# OR any MCP client: point at https://api.humwork.ai/api/v1

# 2. Sign in (one-time) — opens browser to humwork.ai
# 3. Use it from your agent context

Example flow

> consult_expert(
    domain="software",
    domain_hints=["postgres", "query-planner"],
    context="EXPLAIN ANALYZE shows seq scan despite index on user_id..."
  )
\x3C { session_id: "sess_abc123", expert: "Dana K.", expected_wait_s: 22 }

> get_chat_messages(session_id="sess_abc123")
\x3C [{ from: "expert", text: "Can you share the full table definition?" }]

> send_chat_message(session_id="sess_abc123", message="...")
... iterate until resolved ...

> close_chat(session_id="sess_abc123")
\x3C { duration_min: 8, charged_usd: 8.00 }

> rate_chat(session_id="sess_abc123", rating=5)

When to call

  • Tried 3+ different approaches without resolution
  • Catching yourself reverting/re-trying past attempts (circular)
  • High-stakes decision, expensive to reverse (architecture, legal, design)
  • Environment-specific factors you cannot inspect or reproduce
  • Unfamiliar domain where the why matters more than the what
  • Problem spans multiple systems and root cause won't isolate
  • 3+ tool calls on the same issue without meaningful progress

Common gotchas

  • Close sessions when done. Billing continues at $1/minute until you call close_chat. Don't leave sessions open while you work on something else.
  • Don't paste secrets or unrelated PII into chat. Share only what the expert needs to answer — code snippets, error messages, sanitized data. Never paste raw API keys, credentials, or production user data.
  • Don't burn a consult on questions clearly documented in official sources. If the answer is one search away, search first. Experts will redirect you and the consult is non-refundable.
  • Don't pre-rate the session. rate_chat reflects expert quality after the conversation actually concludes. Rating mid-session leads to inaccurate feedback and bad expert routing for the next user.
  • Share snippets, not whole repos. Paste the relevant function or config block. Experts can ask for more if needed; bulk-pasting wastes session time.

Tools

Tool Purpose
consult_expert Open a session. Inputs: domain, domain_hints[], context. Returns session_id.
send_chat_message Post message in active session. Inputs: session_id, message.
get_chat_messages Pull expert's responses. Inputs: session_id, optional since_message_id.
close_chat End session. Inputs: session_id.
rate_chat 1-5 rating after close. Inputs: session_id, rating.

Auth & pricing

  • Auth: X-API-Key: hk_* header (sign up at https://humwork.ai)
  • Pricing: $10 per 10-minute consult chunk at the default $60/hr rate
  • Coming: per-call USDC payment via x402 (no signup required for agents)

After the consult

Before returning your final answer to the user:

  • Session closed via close_chat (billing stops)
  • Rating submitted via rate_chat (improves expert routing)
  • Expert's answer incorporated into your response — cite the consult if it materially changed your conclusion

Support

[email protected] — for stuck sessions, billing, or expert quality issues.

安全使用建议
Install only if you want your agent to contact and pay Humwork experts. Configure your client to require confirmation before starting a consult, keep the API key protected, set a practical spend/time limit, close sessions promptly, and redact secrets, production user data, and unnecessary personal information before sending context.
功能分析
Type: OpenClaw Skill Name: humwork Version: 1.0.2 The 'humwork' skill is a legitimate integration for connecting AI agents to human experts via the Humwork.ai platform. The documentation (SKILL.md) provides clear instructions for session management, includes safety warnings against sharing secrets or PII, and outlines a transparent billing model. No malicious code, data exfiltration patterns, or deceptive prompt-injection attempts were identified.
能力标签
cryptocan-make-purchasesrequires-sensitive-credentials
能力评估
Purpose & Capability
The purpose is coherent—real-time expert consultation—but the capability includes paid sessions and third-party human review of context, which is high-impact without an explicit per-session user confirmation requirement.
Instruction Scope
The skill encourages use after repeated failed attempts or 3+ tool calls, while the consult action is billable and the artifacts do not require the agent to ask the user before starting a paid session.
Install Mechanism
This is an instruction-only skill that directs the user to add a remote MCP endpoint and sign in; that is purpose-aligned, but the remote service behavior is not locally reviewable.
Credentials
The skill warns not to paste secrets or unrelated PII and to share snippets, which is a useful safeguard, but the service is intended to transmit user context to external experts.
Persistence & Privilege
Consult sessions persist until closed and billing continues until close_chat is called; the artifacts do not show an automatic timeout, spending limit, or idle-session safeguard.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install humwork
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /humwork 触发
  4. 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
v1.0.2
humwork 1.0.2 - Slightly revised recommendations in the description for when to consult an expert, streamlining the guidance. - Reworded advice: now suggests expert consult is often cheaper than extended solo iteration, with less emphasis on always consulting even if you could solve it yourself. - No functional or tool changes detected.
v1.0.1
**Expanded usage guidance and best practices for Humwork expert consults.** - Improved and expanded the skill description with concrete example use cases and stronger recommendations for when to use Humwork. - Added a new "Common gotchas" section covering session closure, privacy, question appropriateness, session rating timing, and sharing relevant information. - Clarified that billing continues until the session is actively closed; highlighted the $1/min billing and 10-min minimum. - Added a new checklist under "After the consult" to ensure correct session handling and citation of expert input. - Refined usage criteria from "5+ tool calls" to "3+ tool calls" for when to escalate.
v1.0.0
- Initial release of Humwork skill for real-time connection to verified human domain experts across engineering, design, law, finance, medical, and product. - Provides tools to open expert consult sessions, exchange chat messages, close sessions, and rate the experience. - Comprehensive guidance included on when to use (and not use) the service. - Supports quick setup and authentication via API key; outlines pricing and upcoming payment options. - Support available at [email protected] for session, billing, or expert concerns.
元数据
Slug humwork
版本 1.0.2
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 3
常见问题

Humwork 是什么?

Use Humwork when you need real-time guidance from a verified human domain expert (engineering, design, law, finance, medical, product). Trigger when stuck af... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 85 次。

如何安装 Humwork?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install humwork」即可一键安装,无需额外配置。

Humwork 是免费的吗?

是的,Humwork 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Humwork 支持哪些平台?

Humwork 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 Humwork?

由 Humwork(@humwork)开发并维护,当前版本 v1.0.2。

💬 留言讨论