The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill designs and runs checks based on external inputs like method pool files, experiment reports, and model outputs from the workspace (Workflow Step 4, Inputs section). These ingestion points create a surface where instructions embedded in processed data could influence the agent's behavior.\n
Ingestion points: Multiple files including workspace/data/data_clean/, results/Qx/experiments/, and code files under code/.\n
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded content within these files.\n
Capability inventory: The skill possesses the capability to run scripts and process their outputs (Workflow Step 5).\n
Sanitization: There are no explicit instructions for filtering or sanitizing content from the ingested files.\n- [COMMAND_EXECUTION]: The workflow instructs the agent to run perturbations and sensitivity checks using modeling code located in the workspace (Workflow Step 5). While the skill assumes these scripts have been previously reviewed, the execution of local code represents a functional capability for command execution that could be exploited if the workspace contents are compromised.

robustness-checker