openviking-memory

Fail

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: HIGHDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses local paths that typically store sensitive configuration and environment variables.
  • Evidence: The documentation refers to reading from ~/.openclaw/openviking.env and ~/.openviking/ov.conf.
  • [REMOTE_CODE_EXECUTION]: The skill documentation includes examples of an insecure remote code execution pattern.
  • Evidence: The command curl -fsSL ... | bash in the Multi-Instance Support section.
  • [COMMAND_EXECUTION]: The skill performs command-line operations to manage the agent and execute setup scripts.
  • Evidence: Use of npx for a setup helper and openclaw for configuration and gateway management.
  • [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by storing and recalling arbitrary user text.
  • Ingestion points: Untrusted data from user messages and the memory_store tool (SKILL.md).
  • Boundary markers: Absent; there are no clear delimiters mentioned to isolate recalled memories.
  • Capability inventory: Uses memory_recall to automatically inject stored data into the prompt context (SKILL.md).
  • Sanitization: Absent; no validation or escaping of memory content is described.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 10, 2026, 11:56 AM