Execution reference: This page specifies the tool protocol for automation agents. For instructions to follow in Studio, use the user guide.
This flow creates an orchestrator project agent that delegates to specialists with veryfront__invoke_agent. The result is a group of project agents, not a Veryfront Code workflow primitive.
Agent-facing flow
- Load the Veryfront router skill:
veryfront__load_skill({ "skillId": "veryfront" }). - Before the first agent mutation, load
veryfront__load_skill({ "skillId": "veryfront", "file": "references/agent-create-agent.md" }). Follow that procedure for each specialist and the orchestrator, including catalog discovery and model selection. - If the request is broad, ask exactly one short
veryfront__form_inputwith one requiredbrieffield, then continue from the submitted brief. - Inspect existing agents plus relevant project tools and integrations before creating new primitives.
- Create or update the specialists first. Use only those the brief needs. For a broad starter, default to two unless the brief calls for a different arrangement.
- Create or update one orchestrator project agent. Its instructions must explicitly call
veryfront__invoke_agent, identify the specialist agent ids it delegates to, pass acontextobject for each delegation, and describe how it validates and combines specialist results. - Use live catalog and project tools to create or update the minimal required agents. Reuse existing agents when the brief asks to extend an existing setup.
- After each new agent is created, call
veryfront__create_agent_avatarwithassign: true. For every created or updated agent, readveryfront__get_agent_source, thenveryfront__get_agent_capabilities, using the same project and source target. Stop if any completion step fails. A saved source alone does not prove effective tools. - Summarize the orchestrator, specialists, delegation flow, assumptions, and verification steps. Include a chat action that tests the orchestrator agent.
Guardrails
- Do not load a removed
create-agentic-workflowsub-skill. - Do not invent integration IDs, tool IDs, provider tool IDs, schedule syntax, or resource IDs.
- Do not create a Veryfront workflow primitive for this starter.
- Do not call
veryfront__create_workfloworupdate_workflow, and do not create aworkflows/file unless the user explicitly asks for the lower-level workflow API. - Do not use placeholder specialist agent ids such as
replace-me,todo, orplaceholder. - Do not create unrelated app, page, route, or chat UI files.
- Complete the orchestrator and all required specialists before reporting success. One helper agent or standalone files do not complete the request.
- Prefer updating existing project primitives when the brief asks to extend an existing agent setup.
- When retrying after a partial run, re-use the submitted brief when available and repair existing incomplete agents instead of asking for the same brief again.
Example pattern
This example uses one orchestrator and two specialists. Import the specialist definitions from the source paths created and verified in the project:
// agents/support-workflow-orchestrator.ts
import { agent } from "veryfront/agent";
import intakeAgent from "./support-intake.ts";
import verificationAgent from "./support-verification.ts";
export default agent({
id: "support-workflow-orchestrator",
name: "Support Workflow Orchestrator",
description: "Coordinate support triage by delegating research and verification to specialist agents.",
model: "anthropic/claude-sonnet-4-6",
maxSteps: 20,
system: `You coordinate support triage work.
Rules:
- Decide whether the request needs intake research, execution, or verification.
- Delegate intake to ${intakeAgent.id} and verification to ${verificationAgent.id}.
- Call veryfront__invoke_agent with a bounded task and a compact context object whenever a specialist should act.
- Combine specialist results into one user-facing answer.
- Escalate conflicting or incomplete specialist results instead of guessing.`,
tools: {
veryfront__invoke_agent: true,
},
suggestions: [
{
title: "Triage support queue",
prompt: "Review new support requests, delegate specialist checks, and return a prioritized handoff.",
},
],
});// intakeAgent and verificationAgent are actual saved, verified specialist results.
const orchestrator = await veryfront__create_agent({
"project_reference": "<project>",
"id": "support-workflow-orchestrator",
"name": "Support Workflow Orchestrator",
"description": "Coordinate support triage by delegating research and verification to specialist agents.",
"max_steps": 20,
"system_prompt": `You coordinate support triage work. Delegate intake to ${intakeAgent.id} and verification to ${verificationAgent.id} using veryfront__invoke_agent. Pass a bounded task and a compact context object on every delegation. Validate and combine the results. Report conflicting or incomplete results instead of guessing.`,
"tool_ids": ["veryfront__invoke_agent"],
"provider_tool_ids": [],
"suggestions": [
{
"title": "Triage support queue",
"prompt": "Review new support requests, delegate specialist checks, and return a prioritized handoff."
}
]
})
await veryfront__create_agent_avatar({
project_reference: "<project>", agent_id: orchestrator.id,
source_target_kind: "project", assign: true
});
await veryfront__get_agent_source({
project_reference: "<project>", agent_id: orchestrator.id, source_target_kind: "project"
});
await veryfront__get_agent_capabilities({
project_reference: "<project>", agent_id: orchestrator.id, source_target_kind: "project"
});Veryfront Code source uses system. Hosted MCP mutations use system_prompt for a full prompt replacement. For partial changes, follow the create-agent procedure’s str_replace flow. The MCP example omits a model override. Resolve any explicit model through veryfront__list_models first, and use the saved specialist IDs in the orchestrator instructions.
The instructions must state when each specialist acts and how to handle conflicting or incomplete results.
Output shape
Return a short implementation summary with the orchestrator agent name and id, source path when available, specialist agent names and ids, tools or integrations used, veryfront__invoke_agent delegation flow, assumptions, and how to test it in chat.