openclaw-relay
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/openclaw_relay.pyleveragessubprocess.runto execute local commands includingnodeandpnpm. This is used to drive the OpenClaw and acpx CLI tools.\n- [REMOTE_CODE_EXECUTION]: The skill provides a transport mechanism to execute commands on remote hosts via SSH. It usesshlex.quoteto sanitize arguments passed to the remote shell.\n- [DATA_EXFILTRATION]: The skill reads session tokens and environment configuration from paths like~/.openclaw/gateway.tokenand~/.config/openclaw/team-handoff.env. These are used locally or relayed via SSH to authenticate with the session gateway.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external data for relaying to other agent sessions.\n - Ingestion points: User input or file content read via
--textor--text-fileflags inscripts/openclaw_relay.py.\n - Boundary markers: The
build_publish_promptfunction uses simple labels (e.g., 'Text:') but lacks strong structural delimiters to isolate user content from instructions.\n - Capability inventory: The skill possesses significant local and remote command execution capabilities (
subprocess.run,ssh).\n - Sanitization: Content is passed to the target session without filtering for embedded instructions, though shell-level injection is prevented by
shlex.quote.
Audit Metadata