The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute various shell commands for its core functionality. This includes using curl to send POST requests to http://localhost:31337/notify for user notifications and standard Unix commands (ls, mkdir, cp, mv, rg, touch) for file system management within the ~/.claude/ environment. It also utilizes the bun runtime to execute local TypeScript tools.- [INDIRECT_PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it is designed to read, analyze, and improve instructions found in other skill files (SKILL.md, Workflows/*.md).
Ingestion points: Reads skill files from ~/.claude/skills/ (e.g., in ValidateSkill.md and ImproveSkill.md).
Boundary markers: None identified in the provided files to separate external content from instructions.
Capability inventory: Shell execution (curl, bun, mkdir, mv, cp, touch, rg, ls).
Sanitization: Employs the SkillHygieneGate.ts tool to verify context sufficiency and scan for sensitive data leakage.- [SAFE]: No malicious patterns such as obfuscation, credential exfiltration, or remote code execution from untrusted sources were detected. The skill's operations are confined to the local environment and are consistent with its documented purpose as a skill management orchestrator.

CreateSkill