The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from a user's repository (source code, test suites, and documentation), which creates a surface for indirect prompt injection.
Ingestion points: The workflow in SKILL.md directs the agent to inspect repository guidance, production code, fixtures, and local contracts (e.g., CONTRACT.md).
Boundary markers: The skill implements a 'Campaign contract' and 'Risk ledger' to bound the agent's scope, though it lacks explicit technical delimiters for sanitizing interpolated repository content.
Capability inventory: The agent is authorized to execute repository-level commands and perform code changes within the defined campaign scope.
Sanitization: There are no instructions for escaping or filtering content retrieved from the repository before it is processed by the agent.
[COMMAND_EXECUTION]: The package contains internal maintenance scripts that utilize the Python subprocess module to perform self-validation.
Evidence: scripts/test_skill.py and scripts/test_validator_regressions.py use subprocess.run to execute the unittest framework.
Context: These executions are confined to the Python interpreter (sys.executable) and target only the local files included within the skill package (fixtures and validator tests) to ensure package integrity before use.

adversarial-test-sweep