usage-view

Fail

Audited by Gen Agent Trust Hub on Apr 26, 2026

Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: The skill reads highly sensitive Claude Code conversation transcripts from ~/.claude/projects/ to extract usage data and prompt content. These files contain private code, chat history, and potential credentials.\n- [COMMAND_EXECUTION]: The background agent executes local Node.js runner scripts with command-line flags constructed from user-provided natural language arguments, creating a potential command injection surface.\n- [PROMPT_INJECTION]: The skill processes untrusted transcript data to generate AI insights, creating a surface for indirect prompt injection from previously stored chat history. 1. Ingestion points: ~/.claude/projects/*.jsonl transcripts. 2. Boundary markers: Absent in AI analysis instructions. 3. Capability inventory: Bash, Read, and WebFetch. 4. Sanitization: No explicit filtering of prompt text extracted from transcripts.\n- [EXTERNAL_DOWNLOADS]: The skill fetches pricing information from the official Anthropic documentation website to ensure accurate cost calculations. This is a legitimate use of a well-known service.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 26, 2026, 11:44 AM
Security Audit — agent-trust-hub — usage-view