knowledge-graph-memory

Pass

Audited by Gen Agent Trust Hub on May 6, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill includes a surface for processing untrusted data from conversation transcripts.
  • Ingestion points: The script reads from ~/.claude/projects/*.jsonl and .entire/logs/entire.log to build the memory vault.
  • Boundary markers: The generated Markdown uses Obsidian callout syntax (> [!quote], > [!info], > [!example]) to separate user, assistant, and tool data.
  • Capability inventory: The script conversation_history.py is restricted to local file reading and writing. It does not perform network operations or execute arbitrary shell commands.
  • Sanitization: It implements a _callout_safe function that utilizes regular expressions to strip XML/HTML tags and convert markdown headers into bold text to prevent injected content from breaking the document structure.
  • [DATA_EXPOSURE]: The skill accesses local transcript files to perform its function.
  • Evidence: The script derives the path to Claude Code transcripts in ~/.claude/projects/ based on the repository root. This access is required for the stated purpose of bridging conversation history and does not target sensitive system credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
May 6, 2026, 03:45 AM
Security Audit — agent-trust-hub — knowledge-graph-memory