The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions (SKILL.md) and the supporting scripts (profile_store.py, validate_skill.py) utilize the agent's shell and file system capabilities to manage documentation and configuration. It accesses the host's CLI and file-handling tools to read project state and write knowledge indices (index.jsonl) and Markdown documents.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent attack surface for indirect prompt injection due to its core function of processing external data.
Ingestion points: Project files, user-provided evidence, and content fetched via the host's browser and search capabilities as part of the distillation workflow.
Boundary markers: The workflow in SKILL.md specifies explicit parsing of the project and target files based on the current request, though it relies on standard LLM boundaries.
Capability inventory: The skill can write files, execute CLI tools, and use browser/search tools provided by the host environment.
Sanitization: The skill employs JSON encoding for data index maintenance to prevent structural errors and requires a specific privacy review step and explicit user authorization before any content is published to external platforms.
[SAFE]: The profile management script (profile_store.py) demonstrates good security practices by explicitly filtering out sensitive fields such as 'api_key', 'token', and 'password' before persisting user-stated values to the shared profile. Additionally, the script uses atomic write operations and restrictive file permissions (0o600) for stored configuration.

lov-distill