The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest implementation specifications and journey definitions from external, potentially untrusted sources to establish validation criteria. This creates an attack surface where malicious instructions embedded in a specification could attempt to influence the agent's behavior during the validation process.
Ingestion points: File SKILL.md describes reading journeys from an "implementation specification or another explicitly accepted source."
Boundary markers: The skill provides mitigation instructions, such as requiring "scoped authority" for consequential actions and confirming that the environment and data under control are the intended targets.
Capability inventory: The agent has the capability to execute shell commands (manuvra), generate and run scripts (disposable driver), and perform UI automation.
Sanitization: There is no explicit requirement for the agent to sanitize or filter the content of the input specifications before processing them.
[DYNAMIC_EXECUTION]: The skill instructions allow the agent to create and execute "disposable drivers," which are described as uncommitted scripts using browser libraries for UI validation. Generating and running code at runtime is a form of dynamic execution, although here it is scoped to a specific testing task.
[COMMAND_EXECUTION]: The skill involves interacting with the host system by executing CLI tools, specifically mentioned as manuvra version and running Manuvra jobs. This demonstrates the capability to perform arbitrary shell operations if the underlying harness allows it.

product-validation