The Agent Skills Directory

Data Ingestion Analysis (Indirect Prompt Injection): The skill identifies and processes external data sources including session transcripts and unanswered user queries. While essential for the performance triage workflow, this activity presents a surface where instructions embedded in user data could potentially influence agent behavior.
Ingestion points: Data is imported via CSV and text files as described in SKILL.md and processed by scripts/cluster_queries.py.
Boundary markers: The skill includes explicit instructions for the user to confirm data is de-identified or authorized for sharing before processing.
Capability inventory: The agent performs file reads and executes a local Python script for data clustering.
Sanitization: The provided clustering script tokenizes and filters text, though it does not provide comprehensive sanitization for complex prompt injection patterns.
Local Command Execution: The skill relies on a local Python script (scripts/cluster_queries.py) to perform lexical clustering of user queries. This involves standard command-line execution (python3) as part of the intended workflow. The script is self-contained and utilizes standard Python libraries without external network dependencies.

agent-performance-triage