openclaw-memory-enhancer

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill ingests untrusted data from chat logs and memory files, which is then interpolated into LLM prompts to provide context.\n
  • Ingestion points: The skill reads Markdown files from ~/.openclaw/workspace/memory/ and JSON/Markdown files from ~/.openclaw/workspace/knowledge-base/ through methods in knowledge_base.py and memory_enhancer.py.\n
  • Boundary markers: The recall_for_prompt method in both memory_enhancer.py and memory_enhancer_edge.py encloses recalled memories within [相关记忆] (Related Memory) delimiters to help distinguish context from instructions.\n
  • Capability inventory: The skill performs local file operations (read/write) within its defined workspace. No network operations, subprocess executions, or dynamic code execution capabilities were identified.\n
  • Sanitization: No specific sanitization or escaping of the retrieved memory content is performed before it is added to the prompt context.\n- [EXTERNAL_DOWNLOADS]: The skill's documentation and code reference standard Python packages, specifically sentence-transformers and numpy, which are used for optional high-accuracy semantic search capabilities.\n- [SAFE]: The skill is designed with a focus on local privacy, storing all memory data on the user's device and performing retrieval without external API calls or telemetry.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 07:00 AM