Tutorial: Agent Collaboration

Learn how to work effectively with AI agents in your Flint.

What We'll Cover

Setting up for agent collaboration
Teaching agents your workspace
Using plugins for structured work
Best practices for human-agent workflow

Step 1: Install the Claude Code Mod

Edit flint.toml:

[flint]
name = "my-project"
type = "flint"

[plugins]
required = [
  "core",
  "living-documents",
  "increments",
  "projects"
]

[mods]
required = ["git", "claude-code"]

flint sync

This installs plugins for structured work and the Claude Code mod for tool integration.

Step 2: Prepare Your Init File

The init file is how agents understand your workspace. Make it comprehensive:

---
id: (auto-generated)
tags:
  - "#system"
---

# My Project

A workspace for [describe your project].

## Purpose

[2-3 sentences about what this Flint is for]

## Current Focus

- Active increment: [Increment 1.0.0 - Foundation](/increment-1-0-0-foundation)
- Key tasks: [Task User Auth](/task-user-auth), [Task API Design](/task-api-design)

## Key Documents

| Document | Purpose |
|----------|---------|
| [(Dashboard) Backlog](/dashboard-backlog) | All planned work |
| [(Dashboard) Increments](/dashboard-increments) | Release tracking |
| [Architecture](/architecture) | System design |

## Plugins

| Plugin | Purpose |
|--------|---------|
| Core | Base conventions and utilities |
| Living Documents | Track document lifecycle |
| Increments | Version releases |
| Projects | Task management |

## Agent Instructions

When working in this Flint:
1. Check current increment for priorities
2. Update task status as you work
3. Follow existing naming conventions
4. Ask before major architectural changes

## Workspace

Codebase linked at [Workspace/Monorepo](/workspace-monorepo)

Step 3: Start an Agent Session

When you start Claude Code (or another AI agent) in your Flint:

First message pattern:

Read Mesh/(System) Flint Init.md to understand this workspace.
Then help me with [your task].

The agent will read the init file and have context for everything that follows.

Step 4: Use Plugins for Structured Work

Creating a Task

Read @Plugins/Projects/init-proj.md and create a new task for
implementing user notifications.

The agent will:

Read the Projects plugin init file
Follow the task template
Create a properly structured task document

Creating an Increment

Read @Plugins/Increments/init-inc.md and create increment 2.0.0
for the notification system.

Using Notepad for Ideas

Read @Plugins/Notepad/init-ntpd.md and capture these ideas
about the authentication flow.

Step 5: Establish Conventions

Document your preferences in the init file:

## Conventions

### Naming
- Tasks: `(Task) Verb Noun.md`
- Concepts: `Noun Phrase.md`
- System files: `(System) Name.md`

### Tags
- `#task` - Actionable items
- `#concept` - Ideas and definitions
- `#ld/living` - Active documents
- `#ld/dead` - Archived documents

### Workflow
1. Tasks start in Backlog
2. Active tasks get `status: active`
3. Done tasks get `status: done`

Agents will follow these conventions when creating documents.

Step 6: Multi-Session Continuity

Agents start fresh each session. To maintain continuity:

Session Notes

Create a session note at start:

# Session 2025-12-18

## Context

Continuing work from yesterday:
- Completed: [Task User Auth](/task-user-auth)
- In progress: [Task API Design](/task-api-design)
- Blocked: Waiting on design review

## Goals Today

1. Finish API endpoint specs
2. Start database migration
3. Update architecture doc

## Notes

[Agent adds notes during session]

Handoff Pattern

End sessions with a summary:

Summarize what we accomplished and what's pending.
Update the relevant dashboards and create a handoff note.

Step 7: Effective Prompting

Good Patterns

Grounded in context:

Looking at [Task API Design](/task-api-design), what's the best approach
for the authentication endpoints?

Plugin-aware:

Use the Increments plugin to update our 1.0.0 increment
with today's completed tasks.

Convention-following:

Create a concept note for "Rate Limiting" following our
existing naming conventions.

Patterns to Avoid

Vague scope:

Help me with the project.  ❌

Context-free:

Write some API documentation.  ❌

Better:

Read [Architecture](/architecture) and [Task API Design](/task-api-design), then write
API documentation for the auth endpoints.  ✓

Agent Capabilities by Plugin

Plugin	What Agents Can Do
Core	Navigate, search, create notes
Living Documents	Track status, manage lifecycle
Increments	Create/update version releases
Projects	Manage tasks and backlog
Notepad	Quick capture of ideas
Plan	High-level project planning

Load the relevant plugin init before asking for plugin-specific work.

Workflow: Daily Agent Session

# Morning
1. "Read the init file and check (Dashboard) Backlog"
2. "What's the status of our current increment?"
3. "Let's work on [task]"

# During work
4. Agent creates/updates notes
5. Agent uses plugins as needed
6. Human reviews, provides feedback

# End of day
7. "Update task statuses based on our work"
8. "Create a session summary"
9. Human commits changes

Troubleshooting

Agent seems confused

Re-read Mesh/(System) Flint Init.md and tell me what you
understand about this workspace.

Agent not following conventions

Look at the existing (Task) files in Mesh/ and follow that
exact format for the new task.

Agent context seems lost

Let's start fresh. Read the init file again, then
check our current increment status.

What You've Learned

Init file importance - The foundation of agent context
Plugin loading - On-demand capability extension
Convention documentation - Teaching agents your preferences
Session patterns - Maintaining continuity across sessions
Effective prompting - Grounded, specific, convention-aware

Next Steps

Module - Plugins & Mods - Deep dive into available plugins
Reference - CLI Commands - All available commands
Guide - Core Concepts - Review mental model

Tutorial: Agent Collaboration

On this page