terrain3d
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external repositories, which creates a potential surface for indirect prompt injection attacks where malicious instructions could be embedded in the files being analyzed.\n
- Ingestion points: The
scripts/preflight.pyscript reads and hashes files from a user-supplied repository path via the--repoargument.\n - Boundary markers: The skill explicitly states that the preflight check is an "offline preflight" and not a runtime or whole-revision verification. It also includes warnings that matches only concern README/LICENSE files.\n
- Capability inventory: The skill has access to
Bash,Write, andEdittools, allowing it to execute scripts and modify files.\n - Sanitization: The provided
scripts/preflight.pyimplements significant security controls, including a blocklist for sensitive paths (e.g.,.ssh,.aws,.env) and usesos.O_NOFOLLOWwithdir_fdto prevent symlink traversal attacks.\n- [COMMAND_EXECUTION]: The documentation references shell commands for cloning repositories, building software with SCons, and modifying system security settings.\n - Evidence:
references/upstream.mdincludes commands such asgit clone,scons, andxattr -d com.apple.quarantine. The skill instructions note that these actions require separate user approval.\n- [EXTERNAL_DOWNLOADS]: The skill documentation includes commands to download source code from external Git repositories.\n - Evidence:
references/upstream.mdcontains commands to clone theTokisanGames/Terrain3Drepository from GitHub.\n- [OBFUSCATION]: The skill uses a URL shortener in its reference documentation, which obscures the final destination of the link.\n - Evidence:
references/upstream.mdcontains the shortened linkhttps://lnkd.in/g2MCk8_Fwhich is used to redirect to the project's repository.
Audit Metadata