The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted data by extracting insights directly from conversation history.
Ingestion points: Conversation context and user-provided insights are captured in SKILL.md (Step 2).
Boundary markers: The skill utilizes structured YAML templates defined in assets/resolution-template.md and enforces strict schema validation through schema.yaml (Step 5).
Capability inventory: The skill is permitted to use Write, Bash, and Grep tools to manage a local knowledge base.
Sanitization: Explicit sanitization rules are defined for filenames in SKILL.md (Step 4), and all metadata fields are validated against strict enums and regex patterns in schema.yaml before documentation occurs.
[COMMAND_EXECUTION]: The skill utilizes shell-based tools for searching and organizing the knowledge base.
Evidence: SKILL.md uses Grep for finding similar insights (Step 3) and mkdir -p via Bash for directory structure management (Step 6).
Mitigation: Risks are mitigated by the fact that command arguments are either static values defined in the schema (categories) or have undergone sanitization (filename slugs).

compound