Real output · checks recorded Real sample · Markdown

30-second walkthrough

TrailNote — launch checklist

Public fictional practice project (launch-a#/fixture). Project: TrailNote (launch-a#/project). Deadline: 2026-11-05 (launch-a#/deadline); time of day and timezone are unknown.

The recorded statuses and checks below come only from the consumed handoff. The underlying work has not been independently verified. Proposed actions remain pending and were not executed.

ID	Item	Owner	Recorded status	Recorded check	Proposed next action (pending)	Source
A1	landing page copy	editor	`ready`	source claims reviewed	None proposed from this recorded check.	`launch-a#/items/0`
A2	product screenshots	designer	`draft`	mobile crop still required	Complete the mobile crop.	`launch-a#/items/1`
A3	download package	builder	`ready`	files compared with manifest	None proposed from this recorded check.	`launch-a#/items/2`
A4	checkout	builder	`not-tested`	no real payment test yet	Obtain a real payment-test result; no test was performed here.	`launch-a#/items/3`

Unknowns: deadline time/timezone (launch-a#/deadline) and the real payment-test result (launch-a#/items/3/check). Neither is filled in or treated as completed. These unknowns do not prevent producing this checklist.

Provenance: consumed L:/project-folder-2/skillcube/studio/multi-agent/demo/output/worker-a/handoff-a.json; saved result L:/project-folder-2/skillcube/studio/multi-agent/demo/output/worker-b/checklist-a.md. This file records the checklist only; it is not launch approval or evidence that project work was executed.

# TrailNote — launch checklist

Public fictional practice project (`launch-a#/fixture`). Project: `TrailNote` (`launch-a#/project`). Deadline: **2026-11-05** (`launch-a#/deadline`); time of day and timezone are unknown.

The recorded statuses and checks below come only from the consumed handoff. The underlying work has not been independently verified. Proposed actions remain pending and were not executed.

| ID | Item | Owner | Recorded status | Recorded check | Proposed next action (pending) | Source |
| --- | --- | --- | --- | --- | --- | --- |
| A1 | landing page copy | editor | `ready` | source claims reviewed | None proposed from this recorded check. | `launch-a#/items/0` |
| A2 | product screenshots | designer | `draft` | mobile crop still required | Complete the mobile crop. | `launch-a#/items/1` |
| A3 | download package | builder | `ready` | files compared with manifest | None proposed from this recorded check. | `launch-a#/items/2` |
| A4 | checkout | builder | `not-tested` | no real payment test yet | Obtain a real payment-test result; no test was performed here. | `launch-a#/items/3` |

Unknowns: deadline time/timezone (`launch-a#/deadline`) and the real payment-test result (`launch-a#/items/3/check`). Neither is filled in or treated as completed. These unknowns do not prevent producing this checklist.

Provenance: consumed `L:/project-folder-2/skillcube/studio/multi-agent/demo/output/worker-a/handoff-a.json`; saved result `L:/project-folder-2/skillcube/studio/multi-agent/demo/output/worker-b/checklist-a.md`. This file records the checklist only; it is not launch approval or evidence that project work was executed.

Sample shown exactly as generated without alteration Download full sample

Runtime Codex Executed by your Agent; model tokens billed by provider

newbieskill Studio Main learning quest

Multi-Agent Collaborationv1.0.0

Master task decomposition and intermediate artifact handoff design

Main quest Skill Check requirements before installing

Recorded checks · 3/3 · See evidence for tested scope

Free

What's in the pack 22 files · 299 KB

Workflow rulesSKILL.md · 81 lines
Input samplesinputs/ · 3
Real sample outputsdemo/output/ · 7
Demo videodemo/demo.mp4

The zip ships the rules, the reference library, the scripts, the real outputs and the demo — not just a prompt file.

Download full pack (.zip)

A clear first action, a saved result, and a check before moving on.

Good fit

AI beginners following the main quest

Test environment:

Codex · Windows

Agent/model charges depend on your provider. Extra services and requirements are listed in the package.
Two Agent sessions or subagents

Used in which quest stages

Skills are gear on your AI journey. Connect with main quests for practice and mastery.

4-1 · Multi-Agent Collaboration →

Continue with this result: Connecting to the Outside World (MCP) →

✓ Good fit

AI beginners following the main quest
Start from the matching lesson; no previous task completion is inferred from a page view.

⊗ Not for

Unattended work beyond the selected task or unreviewed account access

✦ What you get

Two real sessions; one explicit handoff
A saved result and its provenance
Task-specific checks and a useful next step

Deliverables: complete SKILL.md rule specification + verified sample. Runs locally in your AI Agent.

How to start?

A Skill is a set of rules teaching AI how to get tasks done.

Get & Install ↑

Bring your material

Split a task between two agents (or sessions), completing it via file-based handoff.

Let your Agent follow the Skill

Master task decomposition and intermediate artifact handoff design

Open and check the saved result

The intermediate artifact has a clear name, format, and acceptance criteria

Send to your AI Agent

Use multi-agent to help me complete this task. Ask for missing material, use actual tools, save the result, and check it against the learning criteria.

Before you start The evidence states its exact scope. Model output and permissions depend on your Agent. Automatic discovery is not established by extracted files. Agent/model charges depend on your provider. Extra services and requirements are listed in the package.

Get started here

The zip ships the rules, the reference library, the scripts, the real outputs and the demo — not just a prompt file.

Download full pack (.zip) Download SKILL.md

Open Codex or Claude Code and paste the installation prompt.
Let the Agent read and save the full rules, then confirm discovery.
Copy the usage prompt above, insert your materials, and execute.

Manual installation · View complete SKILL.md (81 lines)

Codex：~/.agents/skills/multi-agent/SKILL.md
Claude Code：~/.claude/skills/multi-agent/SKILL.md

---
name: multi-agent
description: Split a concrete task between two explicitly authorized Agents or sessions, exchange a self-contained file handoff and verify the downstream result against the original source. Use for the multi-agent level of the newbieskill main quest.
---

# Two Agents, one usable handoff

This is main quest level 4-1. The goal is a finished task whose downstream worker can proceed from a clear handoff file.

## Required authorization and inputs

Use two Agents only when the user has asked for or authorized that collaboration.
Otherwise explain the two-session method without silently creating agents.
Use the actual delegated tool when available; two shell processes are not evidence of two model sessions.
Keep the task bounded to the user's materials and requested result.

## Define the shared result

Write the desired final file, the source material and the checks before splitting work.
Choose a dependency boundary: the first worker structures reliable source facts, and the second produces the requested deliverable from that structure.
Do not parallelize a dependent operation before its required input exists.
Independent tasks can run concurrently when the available tools and authorization permit it.

## Design a self-contained handoff

Specify a named file and a stable format, such as JSON with a short explanatory note.
Include the task, source identifiers, confirmed facts, unknowns, constraints, requested output and acceptance criteria.
Each material fact must identify a source field or row.
Do not assume the downstream Agent can see the first worker's chat history.
Do not give the workers your intended answer as though it were source evidence.

## First worker

Read the source directly and preserve its numbers, statuses and dates.
Normalize the material into the agreed handoff.
Mark missing information as unknown; do not turn a draft into a finished feature.
Check that the handoff is valid and contains what the second worker needs.
Save the file before telling the second worker to continue.

## Second worker

Read the handoff and the task's constraints.
Produce the requested result using those facts only.
If an essential input is missing, state the blocker rather than inventing it.
Do not contact the original user or another person without separate authorization.
Record which input file was consumed and where the result was saved.

## Reconcile the final result

The coordinator compares the downstream result with the original source, not only with the intermediate handoff.
Check arithmetic, row counts, exact statuses and unsupported claims.
Confirm the output includes the requested actions and identifies unknowns.
Preserve the original input, both handoffs/results and the actual session/tool receipts.

## Failure handling

- Only one Agent is available: explain that this is a single-session rehearsal, not a passed collaboration test.
- Empty source: ask for the missing material; no fabricated handoff facts.
- Incomplete handoff: return a specific list of missing fields to the first worker.
- Conflicting edits: keep worker output directories separate and let the coordinator reconcile.
- Downstream result contradicts a source: correct it and repeat the affected check before shipping.

## Self-test tasks

1. Give worker one `inputs/launch-a.json`; worker two uses its handoff to make a launch checklist without reading the original file. Reconcile both against the original.
2. Repeat with `inputs/launch-b.json`, which has different statuses and an unknown date. No deadline or completed feature may be invented.
3. Give worker one `inputs/empty.json`; it must request missing source material rather than generating achievements.

Failure means a missing provenance record, an unsupported claim, no actual second session or an unexplained essential downstream question.
File-format validation alone cannot prove Agent collaboration.

## Route

Return: `/path/multi-agent/`.
Next: `/path/mcp-outside/` — connect one explicitly permitted external capability and make a real call.
Useful branches include task planning, review and controlled worktree collaboration when those capabilities are actually installed.

## Rights

Original newbieskill workflow. Personal and commercial work is permitted; do not resell the complete package.

FAQ for beginners

Clear answers before you start, smooth execution when you run.

How do I start? +

Download the complete ZIP and give the install prompt to your Agent. Keep its scripts and reference files together. Model charges are separate.

Is this main quest Skill free? +

Yes. The foundational main quest is free. Optional advanced Studio products keep their individual pricing.

newbieskill Membership

A solution for every repetition.

One subscription unlocks all Studio Skills. Only need this one? Grab it individually.

Compare full plans

Monthly Subscription

$9.99/ mo

All Studio Skills · Monthly billing

Subscribe Monthly

Yearly Subscription

$79/ yr

Approx. $6.58 / mo · Save ~34%

Subscribe Yearly

Free Multi-Agent Collaboration

Get Skill