The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external REST APIs without explicit sanitization or boundary markers, making it vulnerable to instructions embedded in API responses.
Ingestion points: references/examples.md uses response.json() and response.text to capture data from remote servers.
Boundary markers: No delimiters or instructions are provided to the agent to disregard commands found within the API payload.
Capability inventory: The skill uses execute_code for script execution and network access.
Sanitization: The instructions do not require sanitizing or validating the API response before reporting it.
[DATA_EXFILTRATION]: The skill instructions and examples explicitly guide the agent to read sensitive credentials from the environment (os.environ) and include them in network requests to external URLs. This capability can be used to exfiltrate local secrets if the agent is directed to a malicious endpoint.
[COMMAND_EXECUTION]: The skill relies on the execute_code tool to run dynamically generated Python scripts to perform HTTP requests and parse results.

rest-api-caller