claude-permissions-optimizer

Pass

Audited by Gen Agent Trust Hub on Mar 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bundled Node.js script to analyze local session history and uses node -e to validate the JSON integrity of settings files after modification.\n- [DATA_EXFILTRATION]: Accesses sensitive local data, specifically Claude Code session transcripts located in ~/.claude/projects/, to extract command history. Note that data is processed locally and not transmitted externally.\n- [PROMPT_INJECTION]: Vulnerable to indirect prompt injection as it processes historical session logs which may contain untrusted data.\n
  • Ingestion points: JSONL session transcripts in ~/.claude/projects/ analyzed by extract-commands.mjs.\n
  • Boundary markers: None present.\n
  • Capability inventory: Modifies the security configuration (settings.json) to alter agent permissions.\n
  • Sanitization: Uses a regex-based classifier in extract-commands.mjs to filter out high-risk or destructive command patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 26, 2026, 03:20 PM