cross-host-rig-commands
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from remote hosts into the agent's context via the OpenRig CLI.
- Ingestion points: Data is ingested through commands like
rig capture,rig ps, andrig send(when receiving remote results), as well as reading remote CLI installation states. - Boundary markers: The documentation mentions the use of JSON envelopes to carry cross-host metadata, providing a structured format for data transport.
- Capability inventory: The skill enables the agent to execute CLI commands (
send,capture,ps,whoami,host add) on remote hosts using SSH and HTTP transports. - Sanitization: No automated sanitization or filtering of remote output is specified; the instructions advise the agent to "inspect the result" and "reconcile its durable effect" manually.
Audit Metadata