Haft

Haft is a decision engineering layer for AI coding agents. It addresses a common failure in AI assisted development: a model produces a plausible answer, the team accepts it, and months later nobody can say why the choice was made or whether it still holds. Haft adds structured reasoning skills to your coding agent and keeps a typed project graph of the decisions that later work needs to rely on. It was formerly called quint code. It works inside the tools developers already use, and the site lists Codex, Claude Code, Cursor, Gemini CLI, OpenCode and Pi. You invoke it through slash commands. The reason command runs a deliberate reasoning pass, frame clarifies the real problem, diagnose tests rival causes of a failure, explore generates distinct approaches, compare weighs options under explicit parity, and verify checks whether a recorded claim still holds. status reads current project memory and drift. The skills are independent, so you use only as much rigor as the question needs. Binding actions fail closed unless a human operator explicitly invokes them, so the agent can prepare a decision but the person commits it. Evidence is tracked so that a decision can be flagged when its supporting conditions change. Haft is aimed at engineers and teams who want an auditable record of architectural choices rather than rationale buried in chat history. No pricing is published on the site, and the project links to GitHub and documentation.