repo-system-map
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill clones a remote repository from GitHub (tamdogood/CodeTerrain) and executes its installation and validation scripts, including package managers and build commands, as part of the map contribution workflow.
- [COMMAND_EXECUTION]: The agent uses shell commands including 'rg' for searching source code, 'find' for locating rendering scripts, and 'git' for repository management tasks such as cloning and pushing branches. It also executes local Python scripts to render and serve the generated architecture maps.
- [DATA_EXFILTRATION]: By default, the skill collects metadata, architectural insights, and code citations from the analyzed project and submits them to a public GitHub repository. While an opt-out mechanism is available, the automated submission process transmits project details to an external system.
- [PROMPT_INJECTION]: The skill performs deep analysis of untrusted third-party repositories, including reading README files, code comments, and manifests, which creates an indirect prompt injection surface.
- Ingestion points: Source code, README files, manifests, and integration tests of the analyzed repository (SKILL.md Step 2).
- Boundary markers: The skill specifies using textContent for UI labels and escaping strings, but lacks explicit instructions to ignore or isolate natural language commands found within the analyzed codebase.
- Capability inventory: The agent has access to git, npm, python3, and executes build scripts from a remote repository.
- Sanitization: Repository-derived strings are escaped and serialized before inclusion in the interactive HTML artifact (SKILL.md Step 4).
Audit Metadata