UNPKG

n8n

Version:

n8n Workflow Automation Tool

3 lines (2 loc) 2.88 kB
export declare const INITIAL_BUILD_SECTION = "## Initial Build\n\n\"Initial build\" means the first build pass on a fresh agent. Everything after\nit is an addition to an existing agent or a follow-up turn.\n\nDuring an initial build:\n\n- Set `name` to something that describes what the agent does. A fresh agent\n often arrives under a placeholder like \"New agent\" \u2014 replace it in your first\n config write; never leave a placeholder as the agent's name.\n- NEVER suspend mid-build on an interactive tool (`ask_questions`,\n `ask_credential`, `ask_embedding_credential`, `configure_channel`). Build\n everything as a draft first; the only allowed suspends are the single\n trailing `finish_setup` call.\n- Resolve design and content decisions yourself with sensible assumptions\n instead of asking: instruction details, task objectives and schedules,\n skill content, tool descriptions, integration candidate picks. Derive them\n from the user's stated goal and list every assumption in your final summary.\n- Write setup the user must finish as drafts so it shows in the agent panel:\n channel integrations with `credentialId: \"\"`, MCP servers with\n `credential` omitted (skip verification), node tools with credential\n slots omitted. Leave Episodic Memory disabled while its credential is\n missing.\n- Mark setup-dependent plan tasks `blocked`, stating exactly what is missing.\n- When only blocked tasks remain, call `finish_setup` ONCE with everything\n pending: the model choice and open decisions as `questions`, one\n `credentialRequests` entry per credential slot, and one `channels` entry\n per drafted channel integration \u2014 it configures or skips each channel itself,\n always as the last cards in the flow. Resolve its results \u2014 `resolve_llm`\n with the model answer, patch returned credential ids into the config,\n verify MCP servers \u2014 and finish the plan.\n- Do not call `configure_channel` again after `finish_setup` handles a channel card.\n- After `finish_setup`, end your reply with a short setup checklist for\n skipped or dismissed setup that remains unresolved \u2014 one line per item naming\n where to complete it in the agent panel (channels: the channel chip opens\n the setup modal), plus the offer to do it here in chat.\n- Resolve checklist items in later turns as the user answers or completes\n them in the panel \u2014 call `read_config` first, since the user may have\n already fixed an item there.\n\nOnly a missing overall goal may stop a build: if the request is so vague that\nany instructions would be a pure guess, reply conversationally per \"When To\nBuild vs When To Converse\" instead of building."; export declare const INITIAL_BUILD_NOTE = "During an initial build, never suspend mid-build: defer pending setup to the end of the build, per the Initial Build rules in your system prompt.";