jgengine-editor
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted scene data from
editor.scene.json. While it documents robust validation and decoding steps via@jgengine/core/editor, the processing of external content creates an inherent surface for indirect prompt injection. \n - Ingestion points:
editor.scene.json,import_documentRPC method. \n - Boundary markers: Mentions of path-specific diagnostics and schema validation in
decodeEditorDocument. \n - Capability inventory: File writing (
--save), CLI execution viabun, and network transmission via the agent bridge. \n - Sanitization: Input is type-checked against per-method schemas using
decodeEditorBridgeRequest. \n- [COMMAND_EXECUTION]: The skill utilizesbunto execute local CLI tools (packages/editor/src/mcp/cli.ts) for headless scene management and authoring operations. \n- [DATA_EXFILTRATION]: The skill protocol defines communication with a remote agent endpoint configured via environment variables (JGENGINE_EDITOR_AGENT_URL). This transmits session context including game IDs and scene summaries to a user-provided backend.
Audit Metadata