autoresearch
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCEDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines a workflow to execute arbitrary evaluator scripts and commands as part of an iterative research loop. This is an intended capability for automated development tasks.
- [INDIRECT_PROMPT_INJECTION]: The agent processes mission specifications and evaluator configurations from the .omc/ directory. If these artifacts are sourced from untrusted environments, they could potentially influence the agent's behavior during the loop. Ingestion points include mission.md and evaluator.json; capabilities include command execution and file system writes; no explicit sanitization or boundary markers are defined in the instructions.
- [PERSISTENCE]: The skill supports periodic execution using platform-native scheduling (cron) to maintain mission progress across sessions, which is a supported feature for stateful research missions.
- [DYNAMIC_EXECUTION]: The workflow involves running experiment cycles and evaluator scripts which may involve dynamic code generation and execution based on mission state.
Audit Metadata