The Agent Skills Directory

[SAFE]: The skill follows a structured and highly constrained process to extract knowledge from pull request metadata and discussions. It correctly uses a specialized CLI tool for data access and maintains clear boundaries by not allowing direct database writes or arbitrary command execution.
[SAFE]: Tool usage is strictly restricted to a controlled CLI interface (knowledge-gate) and read-only repository operations, preventing unauthorized data access or modification.
[SAFE]: The skill identifies a surface for Indirect Prompt Injection as it processes untrusted PR data. Ingestion points: Evidence Bundle changed_files and evidence fields. Boundary markers: None explicitly defined in the data schema. Capability inventory: knowledge-gate CLI calls and selective git diff inspection. Sanitization: The skill implements robust mitigation through its 'Conservative extraction principle' and 'Explicitly stated agreement' requirements, which require confirmed textual agreement before information is extracted, significantly reducing the risk of malicious instruction execution.

extract-candidates