audit-openclaw-security
Warn
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The shell script scripts/collect_openclaw_audit.sh attempts to run host firewall diagnostics using sudo -n, which requires administrative privileges to access network configuration data.
- [COMMAND_EXECUTION]: The skill executes various system diagnostic commands such as lsof, ss, netstat, and docker ps to gather environment metadata during the audit process.
- [DYNAMIC_EXECUTION]: The Python script scripts/redact_openclaw_config.py employs import to dynamically detect and load JSON5 libraries, facilitating the parsing of non-standard configuration files.
- [INDIRECT_PROMPT_INJECTION]: The skill features a data ingestion surface where system command outputs are processed into a Markdown report by scripts/render_report.py. Mandatory Evidence Chain: 1. Ingestion: The script reads diagnostic files containing output from system and OpenClaw CLI commands. 2. Boundaries: The rendered report does not use specific delimiters or instructions to isolate ingested data from agent commands. 3. Capabilities: The skill can execute shell scripts and perform local filesystem operations. 4. Sanitization: It redacts secrets using regex patterns but does not implement specific filtering for prompt injection content in the diagnostic data.
Audit Metadata