claude-code-guide

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (LOW): The skill documents an attack surface for indirect prompt injection by instructing the agent on how to fetch and process untrusted external data.
  • Ingestion points: The guide demonstrates usage of web_fetch for URL content and window.fs.readFile for local file analysis.
  • Boundary markers: Examples lack explicit delimiters or 'ignore' instructions for the content being processed.
  • Capability inventory: The documented tool has broad capabilities including bash command execution, file writing (Write, Edit), and network requests (fetch).
  • Sanitization: While a 'Security Considerations' section exists recommending code review and least privilege, the provided code snippets do not implement sanitization or validation of external input.
  • Data Exposure & Exfiltration (SAFE): The documentation mentions sensitive file paths such as ~/.config/claude/mcp_config.json and .env. However, these are referenced solely for configuration and educational purposes; no scripts are provided to automate the exfiltration of these files.
  • Dynamic Execution (SAFE): The skill describes the use of a REPL environment (JavaScript runtime) and MCP servers. These are standard features of the platform being documented and are not used here for malicious obfuscation or hidden execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 05:10 PM