orca-replay
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill enables the execution of shell commands through the orca_replay and orca_compare tools. These tools are designed to re-run commands captured in historical agent recordings or to run a user-provided verification script during model comparisons.
- [DATA_EXFILTRATION]: The orca_compare tool is designed to upload recording data, which may contain source code, prompts, and configuration, to external AI model providers to evaluate alternative model behaviors.
- [INDIRECT_PROMPT_INJECTION]: The skill processes recorded agent traces stored in the .orca/runs directory. These traces constitute an attack surface where malicious data or instructions embedded in a previous run's logs could influence the agent's current session during analysis or replay.
Audit Metadata