memory-fabric

Pass

Audited by Gen Agent Trust Hub on Mar 26, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill is focused on providing a logic layer for knowledge graph orchestration and does not contain any malicious code or instructions.
  • [EXTERNAL_DOWNLOADS]: The skill documentation references the @anthropic/memory-mcp-server package, which is a trusted dependency provided by Anthropic.
  • [PROMPT_INJECTION]: The skill processes user-supplied natural language to extract entities and relations, which is an inherent indirect prompt injection surface.
  • Ingestion points: User queries are ingested in Step 1 of the Unified Search Workflow in SKILL.md and in the Entity Extraction algorithm in references/entity-extraction.md.
  • Boundary markers: The skill relies on natural language parsing and regex patterns rather than explicit delimiters for untrusted data.
  • Capability inventory: The skill has the ability to create, search, and delete entities and relations in the knowledge graph using tools like mcp__memory__create_entities and mcp__memory__delete_entities.
  • Sanitization: The skill significantly mitigates risk by enforcing graph consistency rules (rules/graph-consistency.md) and requiring explicit user confirmation before any pruning or deletion of data (rules/stale-node-detection.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 26, 2026, 10:46 AM