pome-author-task
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined descriptions and existing task data to generate new instructions for other agents. This creates a surface where external input could influence the behavior of the examinee agents.
- Ingestion points: User interviews and task metadata retrieved from the
list_taskstool. - Boundary markers: Generated tasks use explicit Markdown headers (e.g.,
## Prompt,## Success Criteria) to separate instructions from configuration and state. - Capability inventory: The skill writes to the local
tasks/directory and publishes to a remote catalog atmcp.pome.shusing thesave_tasktool. - Sanitization: The skill prioritizes structured
[code]criteria that bind to validated predicates, reducing reliance on unvalidated natural language judge markers. - [COMMAND_EXECUTION]: The skill documentation and instructions describe the use of the
pomeCLI for linting and validating task files (e.g.,pome checks lint,pome compile-seeds). These are standard operations within the authoring workflow for this platform.
Audit Metadata