The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from multiple external sources, which constitutes an attack surface for indirect instructions.\n
Ingestion points: SKILL.md directs the agent to process code, documents, data, website content, and search results as primary material.\n
Boundary markers: The instructions include an explicit "Authority boundary" section (lines 18-20) that mandates treating all gathered content as untrusted evidence rather than instructions.\n
Capability inventory: The skill possesses capabilities for reading files, conducting web searches, and executing local scripts or tests.\n
Sanitization: No specific content sanitization or escaping mechanisms are described to mitigate the risk of instructions embedded within the processed data.\n- [DYNAMIC_EXECUTION]: The skill requires the agent to generate and run scripts to perform data verification and calculations.\n
Evidence: SKILL.md (lines 40-45) requires that all numerical figures in the verdict must be generated by printed script output, necessitating the creation and execution of verification logic.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves methodology and configuration documents from a remote repository.\n
Evidence: It fetches references such as the evidence ladder and codebase design guidelines from the author's official GitHub repository (csark0812/toolbox).

probe