openclaw-audit-watchdog
Pass
Audited by Gen Agent Trust Hub on Apr 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
openclawsecurity CLI and internal helper scripts to perform audits, format reports, and manage cron jobs.\n- [DATA_EXFILTRATION]: Audit reports containing environment security status are delivered to user-configured external endpoints via DM channels (Slack/Telegram) and SMTP relays.\n- [EXTERNAL_DOWNLOADS]: The skill supports automated updates viagit pulland installation via direct download from the official vendor repository on GitHub.\n- [PROMPT_INJECTION]: The skill ingests findings from external agents during audits and interpolates them into reports processed by the AI agent for delivery, creating a standard indirect prompt injection surface.\n - Ingestion points: JSON output from
openclaw security auditcontaining data from scanned third-party skills.\n - Boundary markers: Reports use markdown headers; untrusted finding content is not wrapped in safety delimiters.\n
- Capability inventory: Shell script execution, network communication (SMTP), and agent messaging tools.\n
- Sanitization: No specific logic exists to filter for instruction-like patterns in audit findings before interpolation into report templates.
Audit Metadata