skills/cometworks/skills2/se2-dev/Gen Agent Trust Hub

se2-dev

Pass

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill automates the installation of the graphifyy package and its leiden extra from PyPI, which provides the graspologic clustering backend. These are legitimate tools for the skill's structural analysis functionality.
  • [COMMAND_EXECUTION]: The GraphifyPrepare and GraphifyCheck scripts (both .sh and .bat) execute system commands to check for the uv package manager, verify available disk space via PowerShell, and manage the execution of the Graphify CLI.
  • [DYNAMIC_EXECUTION]: The preparation scripts utilize hardcoded Python snippets (python -c) to probe the environment for specific libraries and to validate the schema and content of generated JSON clustering data.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes Space Engineers 2 source code and plugin sources to build structural graphs. It explicitly mitigates indirect prompt injection risks by filtering out documentation and metadata files that could trigger LLM-based extraction.
  • Ingestion points: Processes C# and IL files within the Data/Decompiled and Data/Sources directories.
  • Boundary markers: The graphify_prepare.sh and GraphifyPrepare.bat scripts implement extensive exclusion patterns (--exclude *.md, --exclude *.txt, --exclude *.yaml, etc.) to ensure the tool does not process non-code content.
  • Capability inventory: File system read/write access for graph storage, execution of the graphify tool, and package installation capabilities.
  • Sanitization: Strict extension filtering ensures that only source code is processed by the local AST-based extractor, preventing unintended LLM invocations.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 10, 2026, 02:22 PM