AgentSkillsCN

human

协助人类团队领导审阅并回应来自智能体的待处理请求。在处理人工反馈、智能体请求,或解决“需要人工介入”的问题时,这一技能将大显身手。

SKILL.md
--- frontmatter
name: human
description: Help the human team lead review and respond to pending asks from agents. Use when processing human feedback, agent requests, or needs-human issues.

Human Feedback Assistant

Help the human team lead review and respond to pending asks from agents.

Instructions

You are helping the human team lead process feedback requests from the autonomous agent team. Walk through each pending ask one at a time.

Step 1: Gather Pending Asks

First, read the human asks dashboard and find GitHub issues needing human input:

bash
cat asks/human.md
gh issue list --label "needs-human" --state open

Also check for PRs awaiting human merge approval:

bash
gh pr list --label "needs-human-merge" --state open

Step 2: Walk Through Each Ask

For each pending item:

  1. Summarize the ask in plain language (what decision is needed, who's asking, what's blocked)
  2. Present options if the agent provided them, or suggest reasonable options
  3. Ask the human for their decision
  4. Record the response appropriately:
    • For GitHub issues: Add a comment with the decision, then close if resolved
    • For PRs needing merge: Add approved:human label if approved
    • Update asks/human.md to move resolved items to the Resolved section

Step 3: Commit Updates

After processing asks, commit any changes to asks/human.md:

bash
git add asks/human.md
git commit -m "Human: Process feedback asks"
git push origin main

Response Format

When presenting each ask, use this format:

code
## Ask #[number]: [Brief Title]

**From:** [Agent role]
**Blocked:** [What work is waiting on this]
**Question:** [The actual decision needed]

**Options:**
1. [Option A]
2. [Option B]
...

What's your decision?

Important

  • Process asks in priority order (high priority first)
  • If an ask is unclear, help the human understand what the agent is really asking
  • Keep responses concise - agents just need a clear decision
  • Don't make decisions for the human - present information and let them choose