The Agent Skills Directory

[SAFE]: The skill body and its associated scripts are free of malicious patterns, obfuscation, or unauthorized system access.
[INDIRECT_PROMPT_INJECTION]: The skill includes a tool to process user-provided experiment manifests in JSON format.
Ingestion points: scripts/validate_ab_testing_contract.py loads external manifest files.
Boundary markers: None, but data is treated as structured input for a validator rather than executable instructions.
Capability inventory: Capabilities are limited to reading the manifest and printing validation results; no network or subprocess capabilities exist.
Sanitization: The validator enforces a strict schema, verifying all types, mandatory fields, and value constraints (e.g., risk levels and runtime bounds).

ab-testing