The Agent Skills Directory

[SAFE]: The skill manages user preferences and records via profile_store.py. This script includes an explicit blocklist for sensitive keys such as 'token', 'secret', and 'password' to prevent the accidental persistence of credentials.
[SAFE]: Verification of automations is performed through verify_automation.py, which connects to the underlying SQLite database in read-only mode, effectively preventing unauthorized data modification during the inspection process.
[INDIRECT_PROMPT_INJECTION]: The skill processes existing automation records to repair or verify scheduled tasks. While this represents a data ingestion surface, the risk is mitigated by instructions that enforce strict schema validation (e.g., cron syntax) and require the agent to report status back for human review. 1. Ingestion points: The scripts/verify_automation.py script reads from the host's native SQLite automation database. 2. Boundary markers: No explicit prompt delimiters are mentioned for the generated automation text. 3. Capability inventory: The skill is authorized to use shell and computer-use tools. 4. Sanitization: The profile_store.py script validates path segments and filters for sensitive keywords before writing to the profile JSON.

lov-yoda-automation