Bonfire
Your Claude Code, opinionated.
Pip-install a runtime and a Claude Code skill. Run /bonfire scan in chat. Name your assistant — your cadre wires itself around the name.
Two commands, then chat.
Install the runtime
One Python package. No daemon, no service, no auth dance.
Register the skill
Drops a Claude Code skill into ~/.claude/skills/bonfire.
$ pip install bonfire-ai # PyPI distribution; Python 3.12+
$ bonfire install-skill # registers ~/.claude/skills/bonfire/SKILL.md
$ export ANTHROPIC_API_KEY=sk-ant-...
Then, in Claude Code, in any project, run /bonfire scan — the cadre wires itself around your repo.
> /bonfire scan
> what do you want to call me?
Nine roles, one batch.
Bonfire wires nine roles around your project. Each plays its role; none impersonate the others. The config file is checked into your repo as bonfire.toml.
See the pipeline: How it works → · Seven in-pipeline roles run in fixed order. Sage runs cross-pipeline; Architect runs as a pre-pipeline bracket. Reference: Pipeline stages →
You name your assistant.
The naming gesture.
You name your assistant; the cadre wires itself around the name. Three built-in personas; or define your own. The persona affects voice, not behavior — a Warrior is still a Warrior.
$ bonfire persona list
falcor (default)
default
minimal
$ bonfire persona set falcor
Most agents vibe-code. Bonfire bakes.
Bonfire packages the discipline of a real engineering studio: structural TDD at the role boundary, audit-driven release gates, cost ledgers per run.
Read about the discipline: /bonfire/docs/discipline.
Four claims.
It learns.
Vault memory. Priority truncation. Tokens compound because the system remembers what worked last time and what didn't.
It competes with itself.
Dual and triple workflows race. Scouts take different lenses on the same ticket. Best output wins. Never a single-shot answer.
It feeds its own fire.
Every resolved ticket feeds the vault. The line heats itself. Session N+1 starts warmer than session N because fuel accumulates.
It gets better the more you use it.
XP progression across six tiers. The system's handling of your codebase sharpens with volume. Not a plateau tool.
Five curated docs.
Apache-2.0 · public.
Built and maintained by CandyFactory.
Bonfire is the methodology this studio uses on its own work, packaged. Issues and PRs welcome on github.com/BonfireAI/bonfire. Releases follow semver; ADRs are public.
Read about the working discipline, the nine-role cadre, and the cost ledger.