The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions.
Ingestion points: Target interface specifications, including OpenAPI 3.1, Protobuf, and configuration schemas, are read from the codebase via file search and read tools (SKILL.md, Step 1).
Boundary markers: There are no explicit instructions for using boundary markers or delimiters when interpolating external data into prompts, though the skill does reference validation against schema contracts.
Capability inventory: The skill is authorized to use execute_command, write_file, create_file, and edit_file to generate and run test suites.
Sanitization: The skill explicitly mentions ASI05 RCE Guard, which instructs the agent to sanitize parameter values before feeding them into execution harnesses.
[DYNAMIC_EXECUTION]: The skill generates and executes code based on the combinatorial matrices it creates.
Evidence: Step 6 and Step 7 describe transforming matrix rows into executable table-driven test cases (e.g., Go subtests, pytest, Vitest) and executing them in a sandbox environment.
[COMMAND_EXECUTION]: The skill executes external tools and generated test suites to perform its primary function.
Evidence: Instructions in Step 4 and Step 7 direct the agent to run PICT/PyPICT binaries and execute the generated test cases using the system's command execution tool.

combinatorial-testing