model-merging
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION] (LOW): The skill defines configuration templates for Mixture of Experts (MoE) merges in references/examples.md which include positive_prompts. These fields serve as an indirect prompt injection surface if populated with untrusted external data, potentially biasing expert routing. (1) Ingestion points: references/examples.md (within the moe-from-merges.yml example). (2) Boundary markers: Absent. (3) Capability inventory: Shell execution of mergekit-yaml and python scripts documented in references/examples.md. (4) Sanitization: Absent.
- [COMMAND_EXECUTION] (LOW): references/examples.md contains several examples of shell commands used to merge models and push them to external repositories. This includes a bash heredoc that dynamically generates and executes a Python script for pushing to the Hugging Face Hub.
- [EXTERNAL_DOWNLOADS] (SAFE): The documentation references standard ecosystem packages including mergekit, transformers, and torch. No untrusted third-party dependencies or malicious remote scripts were identified.
Audit Metadata