Skills
Navigate
skills/waltersumbon/claude-agent-sdk-skill/claude-agent-sdk-typescript/Gen Agent Trust Hub

claude-agent-sdk-typescript

Fail

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides instructions for installing the Claude Code CLI using a piped shell command: curl -fsSL https://claude.ai/install.sh | bash. This is the official installation method for the service.
  • [EXTERNAL_DOWNLOADS]: The skill references and encourages the installation of the @anthropic-ai/claude-agent-sdk package and the @modelcontextprotocol/server-github MCP server from public registries.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides numerous templates and architectural patterns that ingest untrusted user data into the agent's prompt context without explicit boundary markers or sanitization.
  • Ingestion points: Untrusted data enters the agent context through the prompt parameter in query() calls and ClaudeSDKClient.query() methods documented in SKILL.md.
  • Boundary markers: The provided code snippets do not demonstrate the use of delimiters or 'ignore embedded instructions' warnings for untrusted input.
  • Capability inventory: The documented agent configuration grants capabilities for file operations (Read, Write, Edit, MultiEdit), command execution (Bash), and network access (WebSearch, WebFetch).
  • Sanitization: The snippets do not include examples of input validation, escaping, or filtering before interpolating data into prompts.
  • [COMMAND_EXECUTION]: The skill documents the use of the Bash tool and subagents with command-line capabilities, which are standard features of the SDK being described.
Recommendations
  • HIGH: Downloads and executes remote code from: https://claude.ai/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 5, 2026, 01:08 PM