Agent Skills for Claude Code: 9arm-skills

9arm-skills is an Agent Skills repository designed to ensure coding agents (such as Claude Code) follow verified procedures rather than simply producing results. It structures the author's practical routines into Markdown-based skill files, making them portable to agents.

Unlike simple guideline documents, each skill provides procedural instructions that are activated only in specific situations (triggers) via YAML frontmatter.

Key Skills and Features:

• debug-mantra: Enhances debugging reliability by enforcing four steps: reproduction, identifying failure paths, hypothesis falsification, and evidence cross-checking. It is specifically designed to require attempting falsification before forming a hypothesis.
• scrutinize: A code review skill that verifies the intent of changes, traces actual code paths, prohibits formal approvals, and delivers strict judgments.
• qwenchance: Detects 'loop' states where the agent repeats the same mistakes or exhausts context, determining when to hand off to the user.
• qwen-agent: Divides work boundaries so that complex judgments are handled by the main model, while simple repetitive tasks are performed by lower-cost models (such as Qwen).

This repository can serve as a useful starting point for developers seeking to improve the reliability of agent outputs, or for teams aiming to structure their agent usage rules.