micro-app-architect
Pass
Audited by Gen Agent Trust Hub on Jun 27, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill architecture establishes a surface for indirect prompt injection by designing applications that ingest data from workspace files and user inputs to drive AI workflows.
- Ingestion points: The generated apps process data from workspace files (e.g., "data/*.json") and user-provided input in the UI.
- Boundary markers: The framework utilizes Tiptap JSON with specific "mention" nodes to distinguish file references from other content.
- Capability inventory: The micro-apps utilize platform APIs including "window.Magic.fs" (file read/write), "window.Magic.llm" (AI interaction), and "window.Magic.agent" (agent dispatching).
- Sanitization: While the skill emphasizes structured JSON for data storage, it does not explicitly define sanitization or escaping procedures for data content before it is interpreted by the LLM.
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands and local scripts within the workspace to facilitate project maintenance and validation.
- The "references/legacy-migration.md" documentation guides the agent to use standard file system commands like "mkdir" and "mv" to migrate skill directories to new paths.
- The generation workflow incorporates the execution of a local Python script ("quick_validate.py") from a companion skill to verify the integrity of the output.
Audit Metadata