The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system commands such as docker ps during the pre-flight phase to verify environment health and service status.
[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection because its operational logic is derived from an external file (plan-file) provided at runtime. A malicious file could contain instructions that lead the agent to perform unauthorized actions.
Ingestion points: the plan-file provided as a primary argument to the skill.
Boundary markers: the instructions lack delimiters or specific warnings to distinguish between test plan data and agent directives.
Capability inventory: the agent is configured with access to powerful tools including docker, curl, mysql, and aws sqs.
Sanitization: there is no evidence of validation or sanitization of the test plan content before the agent executes the steps.

executing-test-plans