The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the deepline CLI tool globally using npm install -g. It specifically provides a fallback installation method targeting a non-standard registry at https://code.deepline.com/api/v2/npm/.
[COMMAND_EXECUTION]: The agent is directed to use the deepline CLI for various operations, including authentication (deepline auth), searching and describing existing code (deepline plays search/describe), validating code (deepline plays check), and publishing/verifying versions (deepline plays publish/get/versions). These commands involve both filesystem access and network communication.
[DYNAMIC_EXECUTION]: The primary function of the skill is to generate TypeScript source code for 'orchestrator' and 'transition' Plays (files ending in .play.ts). These generated artifacts are then published to a live environment and executed as state machines, representing a full code-generation-to-execution pipeline.
[INDIRECT_PROMPT_INJECTION]: The engines created by the skill are designed to process untrusted data from 'Customer DB' tables within a state machine workflow.
Ingestion points: Data enters the system via the EngineInput interface, which includes a stable key and a generic payload object containing arbitrary data from database rows.
Boundary markers: While the generated code uses a structured switch statement for state transitions, the skill instructions do not mandate explicit sanitization or validation of the payload data before it is processed by transition Plays.
Capability inventory: The generated engine has significant capabilities, including performing arbitrary database queries via ctx.customerDb.query and invoking other logic through ctx.runPlay.
Sanitization: The skill relies on 'inferring' schemas and routine names from user requests, which creates a surface where crafted user instructions could lead to the generation of logic that mishandles external data.

deepline-engine