se2-dev-plugin
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The preparation scripts (
Prepare.batandprepare.sh) download theuvpackage manager from Astral's official site (astral.sh) and thebusybox.exeutility from its primary Windows port site (frippery.org). These are standard tools for the skill's functionality. - [COMMAND_EXECUTION]: The skill makes extensive use of command-line tools including
gitfor repository management anduvfor Python environment execution. The Python scripts (git_utils.py,download_plugin_source.py) use thesubprocessmodule to invoke these commands with controlled arguments. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it downloads C# source code from external GitHub repositories into the
Data/Sources/directory. This content is then indexed and intended to be read by the AI agent. Maliciously crafted comments or documentation within those external plugins could attempt to influence the agent's behavior. - Ingestion points: External C# source files and repository metadata downloaded to
Data/Sources/bydownload_plugin_source.py. - Boundary markers: None identified; the agent reads raw file contents through standard shell utilities.
- Capability inventory: The agent has access to
Bashcommands,dotnet buildfor compilation, anduv runfor executing Python scripts. - Sanitization: No explicit sanitization or instruction-filtering is applied to the content of the downloaded source code before the agent processes it.
- [DYNAMIC_EXECUTION]: The skill uses the
tree-sitterlibrary to parse and index C# code at runtime. It also provides guidance fordotnet buildoperations to compile the developed plugins.
Audit Metadata