product-validation
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external specifications and live application UIs while holding significant capabilities like computer use and CLI execution.\n
- Ingestion points: External implementation specifications, accepted source documents, and live application interfaces (web pages, APIs, CLI outputs) during the validation process (documented in
SKILL.md).\n - Boundary markers: The instructions include logical constraints such as treating implementation contracts as 'read-only' and mutating only 'authorized validation state,' though these do not provide technical isolation from adversarial content (documented in
SKILL.md).\n - Capability inventory: Execution of the
manuvraCLI tool, as well as native 'computer use' and 'browser use' tools for interacting with the operating system and web browsers (documented inSKILL.md).\n - Sanitization: The skill lacks explicit sanitization or filtering logic for the data ingested from the product under test or the external specifications.\n- [COMMAND_EXECUTION]: The skill invokes the
manuvraCLI tool to run validation jobs.\n - Evidence:
SKILL.mddirects the agent to check the CLI version and executemanuvra runusing an authored JSON job.
Audit Metadata