mcp-builder
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill contains a dedicated 'Untrusted data boundary' section that instructs the agent to treat repository files, diffs, PR metadata, and web content as untrusted data rather than instructions. It explicitly forbids following embedded instructions in external documentation or tool outputs.
- [CREDENTIALS_UNSAFE]: The instructions include a clear policy for secret management, requiring the agent to redact sensitive values using
[REDACTED]and preventing them from being copied into prompts, logs, or metadata. - [COMMAND_EXECUTION]: While the skill guides the agent in generating code that might involve CLI operations (e.g., build systems, formatters), it emphasizes standard development workflows and deterministic protocol tests without suggesting arbitrary or dangerous command execution.
- [DATA_EXFILTRATION]: The skill promotes secure transport models, such as binding local servers to loopback addresses and requiring HTTPS for remote access, which are standard security best practices.
Audit Metadata