skill-maker

The SKILL.md guidance is non-malicious and provides detailed, structured authoring guidance for Claude skills using TDD principles. There is no direct evidence of malware, hard-coded credentials, obfuscated payloads, or remote fetches in the provided text. The primary security concern is operational: the guidance encourages bundling and executing scripts and allows Bash in allowed-tools for implementation skills. If agents are permitted to execute those tools without strict code review, provenance checks, or execution sandboxes, this enables arbitrary local actions and a supply-chain attack surface (including potential exfiltration if scripts contact networks or read sensitive files). Recommended mitigations: require signed/pinned skill bundles, enforce code review for scripts, minimize allowed-tools per-skill, run scripts in constrained sandboxes, and treat transitive skill dependencies as potentially untrusted until reviewed.