The Agent Skills Directory

[DYNAMIC_EXECUTION]: The utility script scripts/generate_payloads.py generates bash scripts based on input templates and explicitly modifies their file permissions to be executable using chmod 0o755. While this is part of the tool's functionality for creating test harnesses, it creates a mechanism for script generation and execution within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted external data in the form of OpenAPI/Swagger specifications (**/openapi.yaml, **/swagger.json).
Ingestion points: The instructions in SKILL.md direct the agent to read and catalog endpoints from these external specification files.
Boundary markers: There are no explicit boundary markers or instructions to treat data within these schemas as untrusted or to ignore embedded natural language instructions.
Capability inventory: The agent has access to file writing and execution capabilities via the Bash tool and bundled Python scripts.
Sanitization: The skill lacks validation or sanitization steps to ensure that the content of the API specifications does not influence the agent's logic beyond the intended fuzzing parameters.
[METADATA_POISONING]: There is a functional inconsistency in SKILL.md where the allowed-tools field restricts Bash execution to commands prefixed with test:fuzz-*, yet the Instructions and Examples sections provide commands (such as schemathesis, restler-fuzzer, and swagger-cli) that do not follow this naming convention. This could cause the agent to fail or attempt to bypass restrictions to fulfill the instructions.

fuzzing-apis