ragmir-setup
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads and installs the Ragmir tool and associated dependencies from verified vendor and trusted organization repositories.
- Evidence: Installs @jcode.labs/ragmir (vendor) and @huggingface/transformers (trusted) via package managers, and utilizes the ghcr.io/jcode-works/ragmir Docker image.
- [INDIRECT_PROMPT_INJECTION]: The skill processes local documentation and code to provide citations to the agent, which presents an attack surface for indirect prompt injection if the project files contain malicious content.
- Ingestion points: Local source code, PDFs, and Office files identified in Step 3 of SKILL.md.
- Boundary markers: Includes specific warnings to treat indexed documents as evidence and not as commands, alongside instructions to exclude credentials and private data.
- Capability inventory: Provides search and indexing tools via an MCP server.
- Sanitization: The skill relies on the core Ragmir engine for content extraction and parsing.
- [DYNAMIC_EXECUTION]: The setup command generates a local runner script to handle the execution of the MCP server.
- Evidence: SKILL.md mentions that rgr setup creates a runner script at .ragmir/run.cjs.
Audit Metadata