The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of third-party skill files (SKILL.md) and agent instruction files (AGENTS.md, CLAUDE.md). This creates an indirect prompt injection surface where a malicious skill being audited could attempt to influence the auditing agent's behavior.
Ingestion points: Processes content from SKILL.md, AGENTS.md, CLAUDE.md, and other referenced files within a skill directory.
Boundary markers: Instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when reading audited files.
Capability inventory: The workflow involves running validation tools (skills-ref), performing file system inventory, and potentially creating symlinks using system commands (mklink or os.symlink).
Sanitization: No sanitization or filtering of the audited content is mentioned before the agent processes it.
[EXTERNAL_DOWNLOADS]: The skill references an external validation tool (skills-ref) located in the agentskills repository on GitHub. This is documented as a reference implementation for specification compliance.
[COMMAND_EXECUTION]: The skill documents standard developer workflows for cross-platform compatibility, including the use of directory junctions or symlinks (mklink /J on Windows or os.symlink on POSIX systems).

improving-skills