The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: As a meta-skill designed to audit other agent skills, this tool ingests and processes untrusted markdown and script files. This creates an attack surface where instructions embedded in the files being audited could attempt to influence the auditing agent's logic.
Ingestion points: The skill reads all content within a user-provided directory, including SKILL.md, scripts, and reference documents.
Boundary markers: The instructions do not prescribe specific delimiters or instruction wrappers for the subagents tasked with reading target skill content.
Capability inventory: The agent reads local files, performs linguistic and structural analysis, and runs a validation script (scripts/validate.py) to check for standard compliance.
Sanitization: The included validation script utilizes yaml.safe_load to prevent code execution via YAML frontmatter, though it does not explicitly filter natural language content for prompt injection patterns.

skill-writing