qiushi-skill

Why we included this project

Agents are getting better at producing output, but many still answer before they check the facts. This skills pack goes after that habit directly: one governing principle and nine working methods, packaged as installable assets for Claude Code, Cursor, Codex, OpenCode, OpenClaw, Hermes, and nanobot, so the same investigative discipline applies no matter which host runs the agent. A zero-dependency CLI copies the bundles into each host's scanned directories, writes a manifest so uninstall removes only what it added, and validates the checkout first. Each method cites the original source texts it draws on, which keeps the reasoning auditable, and the entry skill loads only the methods relevant to the current task instead of dumping all nine at once. It is a practical behavioral upgrade for agents that already produce output but need more rigor, not a framework to learn.