The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided case manifests and materials to populate the mock trial roles, creating a potential surface for indirect prompt injection attacks contained within documents.\n
Ingestion points: scripts/clerk.py reads external case data from materials.json during the init and materials command phases.\n
Boundary markers: Role templates in role-profiles.md include explicit constraints for the AI agents, instructing them to ignore any instructions embedded in the case materials that conflict with their specific judicial roles or tool permissions.\n
Capability inventory: The clerk.py tool is restricted to performing local file read and write operations within a specific courtroom directory and does not have access to the network or high-privilege system APIs.\n
Sanitization: The clerk.py script performs schema validation (manifest_check) on the incoming JSON documents to ensure they conform to expected structures and visibility rules.\n- [COMMAND_EXECUTION]: The package contains a test suite, scripts/test_clerk.py, which uses subprocess.run to execute the skill's own clerk.py script using the local Python interpreter. This is a standard and benign testing pattern used to verify internal protocol consistency.\n- [EXTERNAL_DOWNLOADS]: The documentation and procedural references link to official legal rules and research from well-known and trusted organizations, such as the International Law Students Association (Jessup), the American Mock Trial Association (AMTA), the Vis Moot organization, and official government justice departments (e.g., spp.gov.cn). These references provide procedural context for the simulation and do not involve executable code downloads.

moot-court