Understanding the MEOW Principle for Breaking Down Work in Gas Town

The MEOW principle transforms large, vague goals into concrete, trackable tasks that autonomous agents can execute without human supervision by decomposing work into atomic units called Beads.

Gas Town (gastownhall/gastown) implements the MEOW (Molecular Expression of Work) principle to translate high-level human intentions into autonomous agent workflows. According to the [docs/glossary.md](https://github.com/gastownhall/gastown/blob/main/docs/glossary.md), this methodology ensures work is decomposed into trackable, atomic units that agents can execute autonomously, supported by Beads, Epics, Formulas, and Molecules. The system relies on Git-backed persistence and structured workflows to maintain state across distributed agent executions.

What Is the MEOW Principle?

The MEOW principle serves dual definitions within the Gas Town ecosystem. As defined in [docs/glossary.md](https://github.com/gastownhall/gastown/blob/main/docs/glossary.md), Molecular Expression of Work refers to the practice of breaking large goals into detailed instructions for agents. Simultaneously, the [README.md](https://github.com/gastownhall/gastown/blob/main/README.md) describes the Mayor-Enhanced Orchestration Workflow (MEOW) as the specific seven-step pattern for applying this decomposition.

This principle addresses the fundamental challenge of autonomous agent coordination: converting ambiguous human intent into discrete, observable, and recoverable work units. By leveraging Beads (atomic units stored in Dolt), Formulas (reusable TOML templates), and Molecules (chained Bead workflows), MEOW creates a durable pipeline where each task maintains persistent state even if individual agents fail.

The 7-Step MEOW Workflow

The Mayor-Enhanced Orchestration Workflow outlined in the [README.md](https://github.com/gastownhall/gastown/blob/main/README.md) provides a concrete implementation pattern:

1. Tell the Mayor – Submit a high-level description of the desired outcome using natural language.
2. Mayor analyzes – The Mayor agent parses the request and decomposes it into discrete, executable tasks.
3. Convoy creation – The Mayor instantiates a convoy, which groups related Beads into a trackable unit.
4. Agent spawning – The Mayor launches appropriate worker agents (Polecats, Crew members, or other specialized agents).
5. Work distribution – Each Bead is slung to an agent's Hook (personal work queue) for execution.
6. Progress monitoring – The system tracks the convoy's status in real-time, reporting state changes back to the Mayor.
7. Completion – The Mayor aggregates individual Bead results and presents a unified summary to the user.

Core Concepts Supporting MEOW

Several architectural components defined in [docs/design/architecture.md](https://github.com/gastownhall/gastown/blob/main/docs/design/architecture.md) enable the MEOW principle to function reliably.

Beads and Dolt Persistence

Beads represent the atomic unit of work in Gas Town. Each Bead persists in Dolt, a Git-backed datastore, ensuring that work state survives agent crashes or network failures. This persistence satisfies the Nondeterministic Idempotence (NDI) principle, allowing agents to resume or retry work without side effects.

Formulas for Reusable Patterns

Formulas are reusable TOML templates that describe common workflow patterns, such as code reviews or feature implementations. As documented in [internal/formula/README.md](https://github.com/gastownhall/gastown/blob/main/internal/formula/README.md), these templates provide the Mayor with predefined structures for generating Beads, ensuring consistency across similar tasks.

Molecules vs. Wisps

The MEOW principle distinguishes between two types of Bead workflows:

• Molecules – Durable, chained Bead workflows that model multi-step processes with persistent intermediate state.
• Wisps – Lightweight, transient Beads designed for ephemeral tasks that do not require long-term persistence.

Hooks and the GUPP Rule

Each agent maintains a Hook, which functions as a personal work queue. When work appears on a Hook, the GUPP (Gotta Use Personal Polecat) rule forces the assigned agent to execute the task. This mechanism ensures automatic work distribution without requiring polling or manual assignment.

Convoys for Work Grouping

As detailed in docs/design/convoy/, convoys serve as the primary vehicle for grouping MEOW-generated work units. A convoy encapsulates related Beads, providing aggregate status tracking and lifecycle management across the distributed workflow.

Practical Implementation Examples

The following examples demonstrate how to apply the MEOW principle using the Gas Town CLI.

Submit a high-level request to the Mayor:

# Describe the goal in natural language

gt mayor attach <<EOF
Create a new feature that adds a "preview" button to the UI and updates the backend API.
EOF

The Mayor automatically decomposes this into a convoy with specific Beads:

# The Mayor executes:

gt convoy create --title "Preview-button feature" \
    --description "Add button + API endpoint" \
    --beads "$(gt formula render feature-preview.toml)"

# Generated convoy contains:

#   • UI mockup (Molecule)

#   • Backend endpoint (Formula)  

#   • Integration tests (Wisp)

Inspect convoy progress and individual Bead status:

gt convoy show <convoy-id>           # Display convoy status

gt bead list --convoy <convoy-id>    # Enumerate all Beads

Distribute work to specific agents:

gt sling --to polecat-42 --bead <bead-id>

Monitor execution in real-time:

gt feed --convoy <convoy-id>    # Live stream of status updates

gt watch --bead <bead-id>        # Follow specific Bead until completion

Summary

The MEOW principle in Gas Town provides a disciplined approach to autonomous work decomposition:

• Molecular Expression of Work converts vague goals into atomic, trackable Beads stored in Dolt.
• The Mayor-Enhanced Orchestration Workflow provides a seven-step pattern from request to completion.
• Convoys group related Beads while Hooks enable automatic agent work distribution.
• Formulas and Molecules ensure reusability and durability across complex workflows.
• Git-backed persistence guarantees Nondeterministic Idempotence, allowing reliable recovery from agent failures.

Frequently Asked Questions

What does MEOW stand for in Gas Town?

MEOW stands for Molecular Expression of Work, which is the overarching principle for decomposing large goals into agent-executable instructions. It also refers to the Mayor-Enhanced Orchestration Workflow, the specific seven-step implementation pattern described in the [README.md](https://github.com/gastownhall/gastown/blob/main/README.md). Both definitions emphasize breaking work into atomic, trackable units that autonomous agents can process without human intervention.

How does the Mayor agent break down work?

The Mayor agent receives natural language requests via gt mayor attach, then analyzes the intent using built-in decomposition logic. It references Formulas from [internal/formula/README.md](https://github.com/gastownhall/gastown/blob/main/internal/formula/README.md) to instantiate appropriate Bead structures, creates a convoy to group them, and uses gt convoy create to generate the workflow. The Mayor then spawns worker agents and slings individual Beads to their respective Hooks for execution.

What is the difference between a Molecule and a Wisp?

A Molecule is a durable, chained workflow of Beads designed for multi-step processes requiring persistent intermediate state, while a Wisp is a lightweight, transient Bead for ephemeral tasks. Molecules maintain Git-backed history in Dolt and support resumption after failures, whereas Wisps execute once and may not persist long-term. The Mayor chooses between these types based on the complexity and durability requirements of the decomposed task.

How does MEOW ensure fault tolerance?

The MEOW principle implements Nondeterministic Idempotence (NDI) by persisting all Beads in Dolt, a Git-backed datastore. If an agent crashes during execution, the Bead remains in its last known state and can be reassigned to another agent via gt sling without duplicating work or corrupting data. This architectural guarantee, documented in [docs/design/architecture.md](https://github.com/gastownhall/gastown/blob/main/docs/design/architecture.md), ensures that partial failures never result in lost work.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}