claude-agent-builder-typescript
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: The skill follows security best practices and is designed for educational and template-driven agent development. It includes explicit instructions for input sanitization and proper credential management.
- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for agents that process external data from untrusted sources (e.g., CSV files, customer intake forms, API responses), creating a potential attack surface. However, this is mitigated by documentation that emphasizes strict input validation, the use of structured data, and clear system prompts.
- Ingestion points: The Agent.run method in assets/agent-template/agent.ts and data-processing tools in references/examples.md accept external content.
- Boundary markers: The documentation emphasizes using clear system prompts and structured outputs.
- Capability inventory: The templates support tools for file system access, shell execution (via Bash tool), and external API interaction.
- Sanitization: references/custom-tools.md explicitly instructs on input sanitization techniques to prevent injection attacks.
- [EXTERNAL_DOWNLOADS]: The skill references and installs standard development dependencies from official registries, including the @anthropic-ai/claude-agent-sdk and other well-known libraries (e.g., Zod, Vitest). These are documented correctly as part of the development workflow.
Audit Metadata