AI product Open source
optim-plans is a human-in-the-loop planning plugin for Claude and Codex. It turns repository-change requests into traceable Markdown plans, asks planning questions one at a time, records decisions, applies reviewer or criticizer refinement passes, and requires explicit scope confirmation before writing versioned plan artifacts. Its five public skills cover small, broad, and high-risk planning, diagnosis before planning, and reference analysis before planning. After approval, it hands the accepted plan back to the current agent session for normal implementation. Version 0.3.0 keeps machine state in the Git common directory and public artifacts in repository documentation, but does not include a separate execution engine, manifest, delegated implementation role, verification role, retry loop, or worktree execution gate.
1 use taken from transcripts — each links to the moment in the video.
Gives Claude and Codex a structured workflow from vague requests to reviewed code with a human gate. It records planning decisions in versioned Markdown, then executes approved work from an immutable manifest in a dedicated Git worktree.
1 in the library.