anthropic-sdk-security-scan
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze user-provided code, which acts as an untrusted data source. This creates a surface for indirect prompt injection where malicious code could attempt to override the auditor's instructions. The skill mitigates this by explicitly instructing the model to use XML boundary markers and treat the content as data only.
- Ingestion points: Analysis targets specified in SKILL.md (files importing Anthropic SDKs).
- Boundary markers: Recommends using tags and clear instructions to separate data from commands.
- Capability inventory: The skill does not possess any risky capabilities like file system access or network requests.
- Sanitization: Not applicable to the skill's instructions, but the skill guides the agent to verify sanitization in the audited code.
- [EXTERNAL_DOWNLOADS]: The skill provides references to official documentation and GitHub repositories for the Anthropic SDK. These are trusted, well-known resources provided for reference purposes.
- References: Links to official Anthropic SDK repositories on GitHub and documentation at docs.claude.com.
Audit Metadata