菜单
Pressure-test a converged, exported Devague spec for blind spots before planning. Use when the user asks to challenge a spec, find missing risks or unknowns, run a blind-spot pass, or proceed from think to spec-to-plan.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/agentculture/reachy-mini-cli --skill challenge命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Find a Reachy Mini Wireless on the local network without a human typing an IP: sweep the LAN, remember which unit is yours, and log into it over SSH. Wraps `reachy wireless` (`reachy/discover/`) — a stdlib-only HTTP probe plus a bounded, capped, deadline-limited subnet sweep — so an agent gets the same discovery a human operator has. Use when the user says "find my reachy", "where is the robot", "what's the robot's IP", "ssh into the robot", "pin the robot's hostname", or when an agent needs the robot's `base_url` before driving it with `--base-url` / `REACHY_BASE_URL`.
Ask the repository's Colleague harness for an independent read-only explore or review, a previewed scoped write, or feedback on a finished work item. Use when the user asks to consult a colleague or when a fresh, independently verifiable perspective would materially improve a non-trivial repo task.
Fan out a converged Devague plan's dependency waves to isolated workers with human approval of the implementation split and test-gated reconciliation. Use when the user asks to assign a plan to a workforce, fan out work, or run approved plan waves in parallel.
| name | challenge |
| description | Pressure-test a converged, exported Devague spec for blind spots before planning. Use when the user asks to challenge a spec, find missing risks or unknowns, run a blind-spot pass, or proceed from think to spec-to-plan. |
Read
../../../.claude/skills/challenge/SKILL.md
completely before taking task actions, then follow it as the canonical operating
procedure.
Route findings only through the deterministic Devague moves named there. Keep agent-origin findings proposed until the user adjudicates them, and never claim that a pass eliminated unknown unknowns.