The Agent Skills Directory

[PROMPT_INJECTION]: The skill analyzes potentially untrusted project data such as CLAUDE.md, source code, and git logs. While this presents a surface for indirect prompt injection, the agent's capabilities are restricted to local file operations, minimizing risk. * Ingestion points: CLAUDE.md, .claude/agents/, src/, design/, tests/, production/, and git history. * Boundary markers: None. * Capability inventory: Read, Glob, Grep, Write. * Sanitization: None.
[COMMAND_EXECUTION]: The skill instructions mention reading git log, which requires command execution in most environments. However, the skill's allowed-tools are strictly limited to file system access, and the operation is used solely for project summarization.

onboard