nav-graph

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (LOW): The skill stores unsanitized user input from corrections and task documents into markdown files and a JSON knowledge graph, which could be used to inject instructions into the agent's long-term memory.\n
  • Ingestion points: Reads from .agent/.user-profile.json and .agent/tasks/TASK-*.md.\n
  • Boundary markers: Absent. The skill does not use delimiters or 'ignore instructions' warnings when storing user-derived content into .agent/knowledge/memories/.\n
  • Capability inventory: Writes and updates files in the .agent/knowledge/ directory and modifies graph.json.\n
  • Sanitization: Lacks escaping or validation for user-provided strings (e.g., summary, details, pattern) before they are interpolated into markdown templates in memory_writer.py.\n- Data Exposure & Exfiltration (SAFE): The skill interacts with local agent data files (.user-profile.json, graph.json) but does not perform any network operations or access sensitive system credentials.\n- Remote Code Execution (SAFE): No remote code execution patterns, external package installations, or dynamic code execution (e.g., eval, exec) were detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:31 PM