The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses ae-cli for discovery and data retrieval from the AE platform. This is the primary intended mechanism for platform interaction and follows a read-only policy by default.
[DYNAMIC_EXECUTION]: The skill includes a local script, scripts/analyze_experiment.py, which is used for deterministic statistical analysis (SRM and group comparisons). The script is self-contained, using only standard math and JSON libraries, and includes convergence limits and safety checks to prevent resource exhaustion.
[INDIRECT_PROMPT_INJECTION]: The skill processes experiment reports and metadata from the platform.
Ingestion points: Experiment configurations, metric definitions, and report results retrieved via ae-cli.
Boundary markers: The diagnostic-playbook.md and decision-framework.md establish explicit "Trust Gates" to validate data reliability before interpretation.
Capability inventory: Uses ae-cli and analyze_experiment.py for analysis; lifecycle actions (rollouts) require separate explicit user confirmation.
Sanitization: External data is processed as structured JSON by the calculation script, minimizing the risk of instruction leakage through metric names or identifiers.

ae-experiment-insight