The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the tdx agent test command to execute agent conversations and evaluate them against defined criteria.
[PROMPT_INJECTION]: Potential for indirect prompt injection as the tool processes user_input and criteria from test.yml files. Malicious content in these files could attempt to influence the agent under test or the judge agent performing the evaluation.
Ingestion points: user_input and criteria in test.yml.
Boundary markers: None explicitly mentioned in the documentation.
Capability inventory: Executes shell commands (tdx), interacts with judge agents, and reads local agent configuration files (agent.yml, prompt.md).
Sanitization: None explicitly documented for the input fields.

agent-test