The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill scripts read from a local database file (tenkai.db) located within the project directory. This data access is necessary for performing the analysis and does not involve any external network connectivity or access to sensitive system files.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays logs from previous agent executions, which constitutes a data ingestion surface.
Ingestion points: The scripts/analyze_patterns.py script reads payload data from the run_events table, which includes arbitrary text such as agent thoughts and tool outputs.
Boundary markers: Untrusted content is printed to the console without specialized delimiters, which could theoretically allow the data to influence the agent's behavior.
Capability inventory: The skill provides utility scripts for analysis but does not grant excessive system permissions.
Sanitization: Content is truncated for readability but otherwise displayed as stored in the database.
[DYNAMIC_EXECUTION]: The scripts/analyze_experiment.py script uses sys.path.append to facilitate the loading of a sibling module (success_determinants.py). This is a standard and benign use of path manipulation for local library management.

experiment-analyst