mreq-graph
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/page.pyusessubprocess.runto callnodeand execute a local validation script (scripts/check.mjs). This process is used solely to verify the correctness of Mermaid diagram source code before it is included in the output HTML. The command uses a list of arguments and avoids shell execution. - [DYNAMIC_EXECUTION]: The Node.js script
scripts/check.mjsutilizes thevmmodule to load and execute thepele.global.jslibrary in a restricted context. This is a legitimate use case for running a diagram rendering engine to perform syntax checks during the build process. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted data from user descriptions and project files. It mitigates injection risks through multiple layers:
- Ingestion points: User input and referenced project files are processed in
SKILL.mdandscripts/page.pyto generate the diagram content and metadata. - Boundary markers: The skill requires an explicit confirmation step where the agent restates its understanding to the user before generating the diagram.
- Capability inventory: Capability is restricted to reading local project files and writing exclusively to the
.mreq/graph/directory within the session root. - Sanitization: The
scripts/page.pyscript performs rigorous sanitization by escaping HTML entities and using regular expressions to filter out dangerous tags such as<script>,<meta>, and event attributes likeonclick. Furthermore, the output HTML is protected by a strict Content-Security-Policy (CSP) that only allows hashed inline scripts, effectively preventing cross-site scripting (XSS) or remote data exfiltration from the generated page.
Audit Metadata