megatron-impact-mapper
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones the MindSpeed repository from 'https://gitcode.com/Ascend/MindSpeed.git' to a local cache directory in the user's home folder. This is a functional requirement for mapping code changes and utilizes the vendor's own infrastructure.
- [COMMAND_EXECUTION]: Python scripts execute 'git' and 'rg' (ripgrep) via subprocess.run calls. These calls use list-based arguments without spawning a shell, which mitigates command injection risks from user or external input.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests Megatron change event data to identify adaptation points in MindSpeed, representing an attack surface for data-driven instructions. * Ingestion points: Implementation units and event descriptions from input JSON payloads. * Boundary markers: The output is strictly defined by a JSON schema, though no specific delimitation is used for ingested terms. * Capability inventory: The skill can read local files and execute specific search and repository management commands. * Sanitization: The skill includes basic sanitization for branch names and handles search terms through standard subprocess argument arrays.
Audit Metadata