The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes content from external build assets, which creates a potential surface for indirect prompt injection if the source data is compromised.
Ingestion points: scripts/audit_policy.py and scripts/get_assets.py retrieve QUALITY_REPORT, FIDELITY_REPORT, and POLICY_DEFINITION assets using AWS APIs.
Boundary markers: The scripts process these reports as structured JSON/dictionaries; however, there are no explicit boundary markers or warnings to the agent regarding untrusted content within these reports.
Capability inventory: The skill uses boto3 to fetch assets and prints findings to the console. It does not possess write or execute capabilities over the external data, limiting the potential impact.
Sanitization: The scripts perform structural validation (e.g., checking for specific keys) but do not sanitize or escape the policy expressions or report text before reporting them to the agent.
[COMMAND_EXECUTION]: The skill instructions provide workflow commands that execute local Python scripts.
SKILL.md directs the user to run audit_policy.py and get_assets.py using uv run. These scripts interact with the environment via argparse and perform network operations using boto3 to communicate with AWS services.

ar-policy-reviewer