The Agent Skills Directory

[SAFE]: The installation process uses a vendor-verified Node.js library (@mvanhorn/printing-press-library) and downloads platform-specific binaries from the author's official GitHub repository. These resources are consistent with the provided vendor context.
[SAFE]: The skill implements a 'Self-Learning Loop' that uses a local SQLite database stored in the user's standard data directory. The CLI logic includes security safeguards, such as a redaction engine in internal/learn/journal.go that identifies and masks sensitive tokens (e.g., API keys, passwords, secrets) before they are recorded in the invocation journal.
[SAFE]: The teach command, which allows the agent to record new query-to-resource mappings, includes a PII (Personally Identifiable Information) scanner. This utility (ScanPII in internal/learn/recall.go) warns the user if their contribution contains email addresses or phone numbers, encouraging safe data practices.
[SAFE]: The Go source code follows security best practices for database interactions. Dynamic SQL generation for table and column management uses strict identifier validation via regular expressions (validIdentifierRE) to prevent SQL injection, and data manipulation is performed using parameterized queries.
[SAFE]: The skill provides an output delivery mechanism supporting webhooks (--deliver webhook:<url>). This is a documented feature for routing command results and is initiated explicitly by the user or agent based on configuration, rather than acting as a hidden exfiltration vector.

pp-learn-loop-example