No. It's five short habit docs and two copy-paste templates. There's nothing to install and no dependency. You adopt the habits and hand your agent the templates.
No. It's plain Markdown an agent reads, so it works with any coding agent and any model — Cursor, Cline, Continue, Aider, an MCP agent, or a local model via Ollama. Claude Code is just one example. See integrations.md.
Yes. Nothing here calls an API or phones home — it's text. Feed the commit contract and memory files to a local model (Llama, Mistral, Qwen) through Ollama and the whole workflow runs on private code that never leaves your machine.
Yes — captured so it actually happens. The reason repos rot under AI assistance isn't that people don't know better; it's that "review the diff", "verify it works", and "don't git add -A" get skipped under speed. Writing the habits down, and giving the agent a contract, turns good intentions into the default path.
No. Each stands alone. The single highest-leverage piece is the commit contract (COMMIT_RULES.template.md) — start there. Add memory, work-trees, and the verify habit as they earn their place.
It's recommended, not required. Signing proves authorship and integrity, which matters more as a project grows or goes public. If you're not ready, keep every other habit — the system still works.
No — that's the point. The habits are cheap (one extra sentence per commit, one folder per experiment) and they remove the expensive events: untangling merged features, purging a leaked secret from history, discovering a "fix" never worked. Fast and clean.
It's the working method behind a set of shipped solo projects — distilled into something reusable. The commit rules, the memory pattern, and the verify habit are the same ones used to keep those repos clean while building fast with agents.
See CONTRIBUTING.md. A habit earns its place by preventing a real, repeatable failure — bring the failure and the fix.