se-dev-script
Pass
Audited by Gen Agent Trust Hub on Feb 26, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
Prepare.batscript usesreg queryto read the Steam installation path from the Windows Registry (HKEY_CURRENT_USER\Software\Valve\Steam). This is a targeted operation used to locate the game's assets.\n- [COMMAND_EXECUTION]: Setup procedures utilizemklink /Jto create directory junctions, linking local and workshop script folders into the skill's environment.\n- [EXTERNAL_DOWNLOADS]: The skill fetches theuvtool installer fromastral.shand thebusybox.exeutility fromfrippery.org. Both are established and well-known sources for these developer tools.\n- [REMOTE_CODE_EXECUTION]: During initial setup, the skill executes a remote PowerShell script fromastral.shto install theuvpackage manager. While this is a common installation pattern for this vendor, it involves executing code directly from the internet.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes C# source code from the Steam Workshop (SteamScripts/). An attacker could embed malicious instructions in workshop content to influence agent behavior when these scripts are indexed or searched. \n - Ingestion points:
index_scripts.pyreads content fromSteamScripts/(Workshop content) andLocalScripts/.\n - Boundary markers: There are no specific boundary markers or instructions to ignore embedded prompts when the agent reads script contents.\n
- Capability inventory: The skill has capabilities to run subprocesses via
uv runand access the local filesystem.\n - Sanitization: No sanitization is applied to the ingested script source before it is processed or displayed in search results.
Audit Metadata