The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads and writes user profile data to a local JSON file (e.g., ~/.config/agent-skills/profile.json). Evidence: The scripts/profile_store.py script manages this data and includes an explicit blacklist (SENSITIVE_PARTS) to prevent the persistence of keys like api_key, password, or token. The script sets restricted file permissions (0o600) to ensure only the current user can access the profile store.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local markdown and JSONL knowledge records, creating a surface for indirect prompt injection. Ingestion points: Files located within user-configured memory and distill directories. Boundary markers: SKILL.md provides explicit 'Do not activate when' conditions to prevent unauthorized behavior. Capability inventory: The skill utilizes subprocess calls to execute local Python scripts and has file-read access to search the knowledge base. Sanitization: Instructions in SKILL.md mandate safe text matching for punctuation and shell characters to mitigate command injection risks.
[COMMAND_EXECUTION]: The skill is configured to run local validation and profile management scripts. Evidence: README.md and references/user-profile.md describe executing python3 scripts/validate_skill.py and python3 scripts/profile_store.py to maintain the skill's state and integrity.

lov-memory-search