lucid-agent-creator
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides a framework for agents to process untrusted data via a global variable.\n
- Ingestion points: JavaScript handlers ingest external data through a global input variable populated from the request body (SKILL.md).\n
- Boundary markers: The documentation references inputSchema and outputSchema for data validation and allowedHosts for network security, but does not implement specific prompt/data delimiters to prevent instruction injection (SKILL.md).\n
- Capability inventory: Handlers can execute fetch requests (if whitelisted) and return JSON-serializable data. The execution environment is an isolated VM sandbox without access to the host file system or Node.js modules (SKILL.md).\n
- Sanitization: The provided examples do not include explicit sanitization of the input data, relying on schema validation and the platform's sandboxed runtime for safety (SKILL.md).\n- [COMMAND_EXECUTION]: The skill uses the create_lucid_agent MCP tool to transmit agent definitions and handler code to the Lucid platform (SKILL.md).\n- [EXTERNAL_DOWNLOADS]: The skill interacts with vendor infrastructure at api.daydreams.systems and facilitator.daydreams.systems and utilizes external libraries including @lucid-agents/payments and viem (SKILL.md).\n- [DYNAMIC_EXECUTION]: The skill involves the generation and subsequent execution of JavaScript code snippets within a platform-managed VM sandbox (SKILL.md).
Audit Metadata