The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data (Parquet files from production traffic or evaluation sets) to generate monitoring reports. This creates an entry point for untrusted data to enter the agent's context.
Ingestion points: The skill instructions and code examples load data using pandas.read_parquet() in files like SKILL.md, monitoring/quote_drift.py, and ci/churn_drift_gate.py.
Boundary markers: The instructions do not define specific delimiters or instructions to treat dataset content as untrusted strings, which could lead to an agent inadvertently following instructions embedded in the data if it reviews the raw values.
Capability inventory: The skill possesses capabilities for reading the filesystem, writing HTML/JSON artifacts, and triggering alerts via placeholder functions like notify_oncall.
Sanitization: No explicit sanitization or input validation logic is present for the values within the ingested DataFrames before they are passed to Evidently's statistical presets.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the evidently package from the Python Package Index (PyPI). This is a standard and expected dependency for the stated purpose of the skill.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands, primarily for environment setup (pip install evidently) and running monitoring scripts (python -m ci.drift_gate).

evidently-monitoring