omh-executor-runtime-readiness
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external user-supplied requests and target contexts, creating an attack surface where malicious data could influence agent behavior.
- Ingestion points: The skill accepts 'user request' and 'target context' as mandatory inputs (SKILL.md).
- Boundary markers: The instructions explicitly mandate separating 'prepared guidance' from 'observed platform, runtime, connector, file, memory, or delivery evidence.'
- Capability inventory: The workflow involves executing CLI tools including 'omh', 'claude', and 'codex' to perform runtime recording and delegation.
- Sanitization: The instructions include showing the full delegate prompt to the user in a fenced code block, which serves as a manual review gate before execution.
- [COMMAND_EXECUTION]: The skill provides templates and specific instructions for running command-line interface tools.
- It includes a command for recording runtime state: 'omh runtime record --skill executor-runtime-readiness --harness executor-runtime-readiness --status started'.
- It details the construction of commands for the 'claude' and 'codex' CLI tools, including session management and output formatting.
- [DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically generate delegation prompts and CLI command arguments at runtime.
- The agent must compose command strings for subagents, incorporating task-specific flags such as '--allowedTools' or '--permission-mode' based on its assessment of the required permissions.
Audit Metadata