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

se2-dev

Warn

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The preparation scripts (GraphifyPrepare.bat, graphify_prepare.sh) are designed to download and install the graphifyy package and its leiden extra from the Python Package Index (PyPI) using uv, pipx, or pip.
  • [COMMAND_EXECUTION]: The skill utilizes several shell commands for system operations:
  • It executes uv, pipx, or pip for package management and installation tasks.
  • It runs the graphify command-line tool to perform code analysis and graph generation.
  • In GraphifyPrepare.bat, a PowerShell command is executed with -ExecutionPolicy Bypass to calculate directory sizes and check free disk space.
  • The graphify install --platform command in both batch and shell scripts interpolates the SE2_DEV_GRAPHIFY_PLATFORM environment variable or user input directly into a shell command, which may be susceptible to command injection if the input source is untrusted.
  • [REMOTE_CODE_EXECUTION]: The skill implements a pattern of downloading external code (from PyPI) and immediately executing it as part of the tool setup. Additionally, graphify_check.sh executes an inline Python script by piping a heredoc string to the Python interpreter to perform JSON validation.
  • [DATA_INGESTION]: The Graphify tool ingests and processes external codebases from Data/Decompiled and Data/Sources. This content is then used to answer natural language queries through the graphify query command.
  • Ingestion points: Code files located in Data/Decompiled and Data/Sources.
  • Boundary markers: None identified; the tool processes the corpus without explicit delimiters or warnings to ignore embedded instructions.
  • Capability inventory: Includes file system operations (rmdir, rm -rf), package installation, and execution of the graphify CLI.
  • Sanitization: No explicit sanitization or filtering of the ingested code content is observed before graph construction or querying.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 25, 2026, 07:53 AM
Security Audit — agent-trust-hub — se2-dev