thorough-checklisted

# Thorough Check skill (thorough-check) ## Core principle **Evidence, not vibes.** Every conclusion must be backed by concrete command output. Better to list 20 verbose ✅/❌ items for the user than to hand-wave "all OK". ### Typical user trigger phrases **English:** - "Double-check everything carefully" - "Are all the edge cases OK?" - "Is the entire call chain consistent?" - "Verify this change won't break anything" - "Re-check thoroughly before you sign off" **中文 (Chinese):** - "仔细反复的检查 确定都没问题是吧" - "各种边界 各种情况 各种修改地方的流程链条 都是 ok 的是吧" - "多次分析 仔细检查" - "改完仔细检查下" - "确定没问题再放过" What the user wants is for **every implicit worry in your head to be made explicit and verified** — not a polite "all good". ## Language policy This skill's **final report** (the check table shown to the user) defaults to **English** and auto-switches to **Chinese** when: - The user's last 3 messages are primarily Chinese, OR - The session history is primarily Chinese, OR - The user explicitly requests Chinese The workflow instructions themselves (this document) stay in English. Commands and code examples are language-neutral. --- ## Multi-agent cross-check When this skill is invoked for an actual code/config verification task, treat that invocation as standing user authorization to use reviewer agents. Start reviewer agents by default whenever the current runtime and tool policy allow subagents. Token cost is not a reason to skip them. The parent agent still owns the final judgment; subagents