omh-context-budget-review
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user inputs, including task descriptions and context sources, to compute budget plans. This constitutes a standard injection surface, though it is limited by the skill's specific focus on managing context metadata rather than executing instructions within the input.
- Ingestion points:
SKILL.md(Required inputs: task or workflow scope, available context sources). - Boundary markers: The skill establishes a 'not-evidence boundary' and requires separating durable facts from volatile session status.
- Capability inventory: Uses the
omhCLI for runtime tracking and state management; it does not utilize high-privilege tools or perform arbitrary network operations. - Sanitization: No specific sanitization or filtering of the task scope input is mentioned.
- [COMMAND_EXECUTION]: The skill utilizes the
omhCLI tool for runtime recording and context plan management (e.g.,omh runtime record,omh context budget-plan prepare). These commands are used for session-specific state management and observability, consistent with the intended platform functionality.
Audit Metadata