The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data provided by the user, such as case facts, evidence, and party statements, which could contain malicious instructions.
Ingestion points: Untrusted data enters the agent context through the analysis of task_context, matter, and proof fields described in the triaging instructions.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious embedded commands within the user-supplied legal materials.
Capability inventory: The skill instructions include the generation of legal documents (文书草稿), procedure matrices, and strategic responses. Notably, it contains a safeguard requiring explicit user authorization before any submission or contact with third parties.
Sanitization: The skill does not implement specific sanitization or filtering logic for the input data before it is interpolated into the reasoning process.

complaint-triage