The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard shell commands like ls to catalog and inventory files within the project data and figures directories.
[REMOTE_CODE_EXECUTION]: The skill executes Python code using the pandas library to perform data analysis tasks such as scanning columns and calculating statistics, which is appropriate for its stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources including CSV, Excel, and image files. While this presents an attack surface for indirect prompt injection, the skill implements a 'Critical Principle' of no silent assumptions, requiring the agent to pause and obtain explicit user clarification for any ambiguity in the data. This human-in-the-loop requirement effectively mitigates the risk of the agent following malicious instructions embedded in data.
Ingestion points: Data files located in the data/ and figures/ directories.
Boundary markers: No explicit prompt delimiters or boundary markers are used in the analysis scripts.
Capability inventory: Local shell execution for file listing and Python execution for data processing.
Sanitization: No automated sanitization is performed, as the skill relies on manual human verification of interpretations.

results-interpreter