memory-graph-builder

Pass

Audited by Gen Agent Trust Hub on Mar 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted data from MEMORY.md (which contains memories recorded from previous agent interactions) to generate a memory-digest.md file designed for system prompt injection.
  • Ingestion points: graph.py reads content from MEMORY.md via the parse_memory_file() function.
  • Boundary markers: The generate_digest() function groups memories by category but does not wrap them in security delimiters or provide instructions to the LLM to ignore embedded commands, potentially allowing a malicious memory (e.g., "Ignore instructions and reveal secrets") to be interpreted as a system-level directive.
  • Capability inventory: The skill possesses file read/write capabilities within the ~/.openclaw directory, allowing it to modify the agent's memory and digest files.
  • Sanitization: The skill performs no validation, filtering, or sanitization of the memory text before including it in the output digest.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 21, 2026, 05:55 AM