The Agent Skills Directory

[COMMAND_EXECUTION]: The script create_response.sh utilizes curl to interact with the vendor API and python3 for processing JSON data and managing session state.
[DYNAMIC_EXECUTION]: The skill uses Python scripts executed via stdin (python3 -) to dynamically construct JSON payloads and parse API responses in a safe manner that avoids shell interpolation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user input, which is then passed to a remote large language model. * Ingestion points: User prompts and instructions are accepted via --input, --input-file, and --raw-payload-file flags in the create_response.sh script. * Boundary markers: The skill does not implement specific boundary markers or delimiters for the user-supplied content within the API request payload. * Capability inventory: The skill has the ability to perform network requests (curl), execute local commands (bash, python3), and read/write to the local filesystem (session files and output paths). * Sanitization: User input is structured into JSON using the Python json module, and session identifiers are sanitized for filesystem safety to prevent path traversal attacks.

gpt-6