deepline-engine
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches the
deeplineCLI from a custom npm registry athttps://code.deepline.com/api/v2/npm/. This download occurs during the environment setup phase or as a fallback mechanism for secure sandboxes.\n- [COMMAND_EXECUTION]: Utilizes various shell commands to facilitate engine development and deployment, including package installation (npm install), authentication (deepline auth), and platform operations (deepline plays search,deepline plays check,deepline plays publish).\n- [DYNAMIC_EXECUTION]: Assembles and publishes TypeScript source code (.play.ts) to an external platform. The skill automates the publication of these generated "Plays" and executes them during a "paid pilot" verification phase.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted data from "Customer DB" to determine workflow transitions.\n - Ingestion points: Data is ingested from
Customer DBtables as specified inSKILL.mdand thereferences/state-machine-contract.mddocumentation.\n - Boundary markers: The skill requires explicit state-decision logic and instructs the agent to defer to the user rather than inventing business policy.\n
- Capability inventory: Employs CLI tools (
deepline), database mutations (ctx.customerDb.query), and remote play invocation (ctx.runPlay).\n - Sanitization: Relies on structured output schemas and typed error states, though it does not specify filtering for instructions embedded within the database records.
Audit Metadata