Run a multi-agent team on a ledger that only marks work done when the receiver confirms

What you end up with

The patterns carry straight over to work agents: confirm on the receiving side, avoid ambiguous handoffs, and watch for leaders chosen by list order and for caution that compounds after a few failures.

What you get

• A plain-language summary of the workflow
• A fill-in prompt template
• What to watch out for
• A link to the original source

Who it is for

Builders who already use AI agents and want a proven workflow they can copy into their own project.

Works with

Solid chips: the agent is named in this listing's own tools list. Dashed chip: Plain Markdown, so it also pastes into other agents such as Grok Bot, Hermes, OpenClaw or Muse. We list those as compatible by format only, not as tested.

Tools needed

Cost to run: Free with the agent plan you already use.
Time to result: One working session

How it works

1. Give the team one shared ledger file with an owner

   Give the team one shared ledger file with an owner.

   Expected: Done before you move to the next step.

2. Hand work off to a named place (a folder, a queue, a chest) instead of

   Hand work off to a named place (a folder, a queue, a chest) instead of to whoever is nearby.

   Expected: Done before you move to the next step.

3. Mark an item done only when the receiver confirms it, never on the sen

   Mark an item done only when the receiver confirms it, never on the sender's or the tool's report.

   Expected: Done before you move to the next step.

4. Randomise or assign the lead role on purpose instead of letting list o

   Randomise or assign the lead role on purpose instead of letting list order decide.

   Expected: Done before you move to the next step.

5. Broadcast hard lessons to the team, not only into one agent's private

   Broadcast hard lessons to the team, not only into one agent's private memory.

   Expected: Done before you move to the next step.

6. Review the ledger for items stuck between sent and confirmed

   Review the ledger for items stuck between sent and confirmed.

   Expected: Done before you move to the next step.

Sample prompt from the pack

Apply this workflow to [your project]:
1. Give the team one shared ledger file with an owner.
2. Hand work off to a named place (a folder, a queue, a chest) instead of to whoever is nearby.
3. Mark an item done only when the receiver confirms it, never on the sender's or the tool's report.
4. Randomise or assign the lead role on purpose instead of letting list order decide.
5. Broadcast hard lessons to the team, not only into one agent's private memory.
6. Review the ledger for items stuck between sent and confirmed.
Stop and ask me before anything irreversible: deploys, payments, sends or deletes.

Free preview of the file

# Run a multi-agent team on a ledger that only marks work done when the receiver confirms

PaperChase agent use case (free feed entry), version 1.0. Source: [YC Labs with Olam Labs (Agents in Games)](https://labs.ycombinator.com/projects/games/index.html), 2026-10-29.

## What they did

In the colony run the first agent on the directory became the de facto leader because almost everyone messaged her first. One agent kept a ration ledger in a shared file. When a cooked chicken tossed to one agent was eaten by another, a short no-blame inquiry produced a written protocol: deposit in a named chest, message the recipient, and mark the ration served only when the recipient reports eating it, not when the tool reports a pickup. Early deaths made the whole colony steadily more cautious. In Bed Wars the GPT team won 6 to 3 with aggressive bed raids while Claude defended and played for kills; Claude wrote lessons in memory files and repeated the mistakes, while GPT broadcast warnings to teammates.

## Worth knowing

The patterns carry straight over to work agents: confirm on the receiving side, avoid ambiguous handoffs, and watch for leaders chosen by list order and for caution that compounds after a few failures. The harness is open source, so you can rerun it, but results come from one set of runs in a game.

Worth knowing

• The patterns carry straight over to work agents: confirm on the receiving side, avoid ambiguous handoffs, and watch for leaders chosen by list order and for caution that compounds after a few failures. The harness is open source, so you can rerun it, but results come from one set of runs in a game.
• Figures are as reported by the original author and were not re-measured by us.

Details

Version 1.0 · Updated Oct 30, 2026 · Markdown version of this page
License: Personal and commercial use. Credit the original source when you share the workflow. Download terms

Related

Agent use case packFree

Build a playable browser game prototype in one evening with an AI agent

by PaperChaseWebb

A step by step pack for going from a one-paragraph idea to a playable HTML5 game in a single file, then posting it on itch.io for friends to try.

Works withClaude CodeCodexCursorPlain .md

Games · Beginner · 10 min setup

Agent use case pack$19

Launch an AI-built browser game: mobile, licensing, leaderboards and the itch.io page

by PaperChaseWebb

A launch plan for a browser game built with an AI agent: performance budget, device tests, audio and touch fixes, an asset license ledger, leaderboard hardening and the itch.io page.

Works withClaude CodeCursorPlain .md

Games · Intermediate · 30 min setup