The Agent Skills Directory

[SAFE]: The harness skill provides a structural methodology for organizing project documentation to prevent context bloat and maintain architectural awareness.
[PROMPT_INJECTION]: The skill features an indirect prompt injection surface inherent to tools that analyze project files.
Ingestion points: commands/audit.md and commands/init.md read project files such as CLAUDE.md and .cursorrules.
Boundary markers: Absent; external data is processed without explicit delimiters.
Capability inventory: None; the skill does not implement network communication, arbitrary command execution, or file writing.
Sanitization: Absent; the skill does not validate or sanitize the content of the project files it reads.
[SAFE]: All external links are to trusted domains, and no suspicious dependencies or obfuscated code were found.

harness