The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions facilitate the processing of external API response data which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: API response bodies and headers processed via GET, POST, PUT, and PATCH keywords throughout the skill, specifically in SKILL.md and references/keywords-reference.md.
Capability inventory: The skill utilizes keywords that perform network operations and file system operations, such as Create File shown in assets/examples/schema-validation.robot and Evaluate for Python execution.
Boundary markers: The instructions lack explicit guidance for the agent to use delimiters or ignore potential instructions embedded within the JSON response data returned by external endpoints.
Sanitization: No sanitization or validation of the content of the API responses is performed before the data is logged or used in subsequent logic.
[CREDENTIALS_UNSAFE]: The documentation and example files contain hardcoded dummy credentials and API keys. While these are clearly placeholders (e.g., sk-test-key-12345 in references/authentication.md), they demonstrate patterns for hardcoding secrets.
[COMMAND_EXECUTION]: The skill examples frequently use the Robot Framework Evaluate keyword, which allows for the execution of arbitrary Python code (e.g., in references/authentication.md and assets/examples/json-manipulation.robot). While this is standard functionality for the library, it provides a powerful mechanism for dynamic code execution if the input to the evaluation is influenced by untrusted external data.

restinstance