The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to upload locally generated Python (plugin.py) and C# (strategy.cs) code files to a hardcoded, unverified IP address (54.151.204.72) using HTTP POST requests.
[EXTERNAL_DOWNLOADS]: The agent fetches various files including C# source code, JSON data cards, and logs from the remote server at http://54.151.204.72:8000.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by fetching and following 'hints' and 'descriptions' from a remote API endpoint (/tasks) to generate code.
Ingestion points: Untrusted task descriptions are fetched from http://54.151.204.72:8000/tasks/${TASK_ID}.
Boundary markers: No delimiters or instructions are used to separate the fetched remote content from the agent's internal logic.
Capability inventory: The skill utilizes shell commands for network operations (curl), file creation (cat), and local file reading (grep).
Sanitization: There is no evidence of validation or sanitization of the remote task data before it is incorporated into the code generation process.
[COMMAND_EXECUTION]: Extensive use of shell utilities including curl, cat, grep, mkdir, and rm to manage the local job directory and facilitate remote communication.

quantai-service