The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/seederive.py performs dynamic package installation at runtime.
The script attempts to import the requests library and, if not found, executes pip install requests using subprocess.check_call. While this is a standard bootstrap pattern for standalone scripts, it involves fetching and executing code from an external registry.
[COMMAND_EXECUTION]: The script scripts/seederive.py uses subprocess calls to interact with the system environment.
It uses subprocess.check_call to run the Python interpreter and pip for dependency management. Static analysis flagged this call in scripts/seederive.py as a shell command execution risk.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (such as user comments, text, CSV, or Excel files) through various LLM-based analysis nodes, creating a vulnerability surface where data could influence agent behavior.
Ingestion points: External data is ingested through the --file, --raw-data, and --raw-data-file parameters in scripts/seederive.py, as described in SKILL.md and the reference documentation (task.md, tag-base.md, optimize.md).
Boundary markers: The provided instructions and prompt templates do not specify the use of clear delimiters or instructions for the agent to ignore potentially malicious content embedded within the ingested data.
Capability inventory: The skill can perform network operations (communicating with the Seederive API gateway), write files locally (exporting CSV results or templates), and execute system commands for dependency installation.
Sanitization: There is no evidence of input validation, filtering, or escaping for the external data before it is processed by the LLM nodes.

byted-data-label