跳到主要内容
知仓学习社ZHICANG

pending-questions

pending-questions,来自 a5c-ai/babysitter 的 agent 技能。

不碰外部(只输出文字)无严重或高危命中a5c-ai/babysitter

它会碰到什么

扫了多少1 个文本文件,3 KB
它会碰到什么不碰外部(只输出文字)
命中总数0 处
命中统计严重 0 · 高 0 · 中 0 · 低 0

这一栏是扫描器报的事实,不是结论。命中多不等于有毒(安全工具、规则库、示例脚本本来就会包含危险写法),命中少也不等于干净。它和你手上的凭据、文件、网络有什么关系,需要你自己看。

技能内容

Pending Questions -- Responder Answering Skill

When to Use

Use this skill when the user is acting as a responder and wants to:

  • check for pending breakpoints
  • monitor for newly routed breakpoints
  • draft and submit an answer with project-specific context

Step 1: Resolve the Responder Identity

Prefer an explicit responder ID from the user. If they do not provide one:

  1. inspect .a5c/responder/ or run list_responders
  2. match the user to an existing responder profile when possible
  3. ask for the responder ID before answering anything

Cache the resolved responderId and display name for the rest of the session.

Step 2: Poll for Pending Breakpoints

Tool: poll_breakpoints

{
  "responderId": "security-responder",
  "waitSeconds": 0
}

For continuous monitoring, raise waitSeconds and repeat the poll after each cycle:

{
  "responderId": "security-responder",
  "waitSeconds": 30
}

Summarize the pending breakpoints, then ask which one to answer if there is more than one.

Step 3: Claim the Breakpoint

Claim the breakpoint before drafting so other responders can see it is being worked:

Tool: claim_breakpoint

{
  "breakpointId": "bp_123",
  "responderId": "security-responder"
}

If the claim fails, tell the user and let them choose a different breakpoint or continue without claiming when the backend does not support it.

Step 4: Enrich Context

Before drafting an answer:

  • read any referenced files from the workspace
  • inspect related code paths and recent changes when they matter
  • summarize the constraints, risks, and likely trade-offs

The goal is to give the responder a short, grounded brief before they approve the draft.

Step 5: Draft the Answer

Prepare a draft that:

  1. leads with the recommendation
  2. explains the reasoning with repo-specific context
  3. calls out caveats or follow-up checks
  4. proposes an honest confidence score
  5. includes relevant file or doc references

Ask the responder whether to submit, revise, change confidence, or skip.

Step 6: Submit the Answer

Tool: answer_breakpoint

{
  "breakpointId": "bp_123",
  "responderId": "security-responder",
  "responderName": "Sam Rivera",
  "text": "Require SameSite cookies and a CSRF token on the form POST.",
  "confidence": 85,
  "references": [
    "packages/adapters/tasks/docs/expert-guide.md",
    "docs/security/csrf.md"
  ]
}

If the workflow uses signed answers, include:

  • sign: true
  • keyFingerprint

Step 7: Confirm Recorded State

After submission, confirm what the backend recorded:

Tool: check_breakpoint_status

{
  "breakpointId": "bp_123"
}

Offer to return to polling for the next breakpoint.

Continuous Mode

When the user asks you to keep watching:

  1. poll with poll_breakpoints
  2. surface new breakpoints as they appear
  3. help draft and submit answers
  4. return to polling until the user stops

For terminal-first workflows, tasks-adapter responder-loop --responder <responderId> --once is the package CLI equivalent of a single poll cycle.

想直接用这个技能?

本站把开放许可(MIT / Apache 等)的技能按仓库打包整理到网盘,点一下转存到你自己的网盘,不用一个个从 GitHub 拉。许可未声明的技能只给原始仓库链接,不打包。