Bonfire
Your AI coding partner forgets everything between conversations. Bonfire fixes that.
claude plugin marketplace add vieko/bonfire
claude plugin install bonfire@vieko
The Problem
AI agents are stateless. Every conversation starts from zero. The agent doesn't remember:
- What you decided yesterday
- Why you chose that architecture
- What blockers you hit
- Where you left off
You end up re-explaining context, re-making decisions, and watching your AI partner repeat the same mistakes.
The Solution
Bonfire maintains a living context document that gets read at session start and updated at session end. Your AI partner picks up exactly where you left off. It's like a saved game for your work.
/bonfire:start → reads context → WORK → /bonfire:end → saves context
That's it. No complex setup. No external services. Just Markdown files in your repo.
Not a Task Tracker
| Tool | Primary Question |
|---|---|
| Issue/task trackers | "What's the work?" |
| Bonfire | "Where are we and what did we decide?" |
Bonfire complements your issue tracker. Use GitHub Issues, Linear, Beads, or Beans for tasks. Use Bonfire for workflow context.
Quick Start
# Install
claude plugin marketplace add vieko/bonfire
claude plugin install bonfire@vieko
# First run scaffolds .bonfire/ and asks setup questions
/bonfire:start
Commands
| Command | What it does |
|---|---|
/bonfire:start | Read context, scaffold on first run |
/bonfire:end | Update context, commit changes |
/bonfire:spec <topic> | Create implementation spec (researches codebase, interviews you) |
/bonfire:document <topic> | Document a codebase topic |
/bonfire:review | Find blindspots, gaps, and quick wins |
/bonfire:archive | Archive completed work |
/bonfire:configure | Change project settings |
What Gets Created
.bonfire/
├── index.md # Living context (the important one)
├── config.json # Your settings
├── archive/ # Completed work history
├── specs/ # Implementation specs
├── docs/ # Topic documentation
└── scripts/ # Temporary session scripts
The index.md is where the magic happens. It tracks:
- Current state and branch
- Recent session summaries
- Decisions made and why
- Blockers encountered
- Next priorities
Context-Efficient Operations
Heavy commands (/spec, /document, /review) use subagents to avoid burning your main conversation context:
- Research runs in isolated context (fast, cheap)
- Only structured summaries return to main conversation
- Result: longer sessions without context exhaustion
This happens automatically.
Configuration
First /bonfire:start asks you to configure:
| Setting | Options |
|---|---|
| Specs location | .bonfire/specs/ or specs/ |
| Docs location | .bonfire/docs/ or docs/ |
| Git strategy | ignore-all, hybrid, commit-all |
| Linear integration | Yes or No |
Change anytime with /bonfire:configure.
Git Strategies
| Strategy | What's tracked | Best for |
|---|---|---|
| ignore-all | Nothing | Solo work, privacy |
| hybrid | docs/, specs/ only | Teams wanting shared docs |
| commit-all | Everything | Full transparency |
Linear Integration
If you use Linear for issue tracking:
- Install Linear MCP
- Enable via
/bonfire:configure - Reference issues by ID:
ENG-123
Bonfire will fetch issue context on start, create issues from review findings, and mark issues Done on archive.
Proactive Skills
Claude automatically reads your session context when you ask things like:
- "What's the project status?"
- "What were we working on?"
- "What decisions have we made?"
And suggests archiving when you merge PRs or mention shipping.
Requirements
- Claude Code CLI
- Git repository
Optional: gh CLI for GitHub integration, Linear MCP for Linear integration.
Recommended: Auto-Formatting Hook
Keep Claude's code formatted with a PostToolUse hook that runs your project's formatter after every write. See PostToolUse Formatter Hook for setup instructions.
Learn More
Blog post: Save Your Progress
Changelog: CHANGELOG.md
License
MIT © Vieko Franetovic