haft

Why we included this project

Coding agents like Claude Code, Codex, and Cursor are good at writing code, but the decisions behind that code usually stay implicit. haft turns problem framing, option comparison, and the evidence behind a decision into durable records stored in the repository, and it flags decisions that no longer match the current spec or runtime state. The same record set is reachable through the agent's own skills and slash commands, through a plain CLI when you want no LLM in the loop, and through an MCP server that any agent can query programmatically. Teams that want an auditable decision trail rather than just code diffs will find this useful, especially when parity between the spec and the implementation has to be enforced rather than assumed. haft is not a coding agent itself; it is the governance layer you put underneath one.