v1 · terminal-first decision ledger

▚ Ladder

Your AI pair surfaces three viable paths every session. You pick one; the rest vanish into scrollback. Ladder keeps them — and hands them back to Claude Code and Codex, rung by rung, without leaving the terminal.

Sign in Open the ledgerDocs
your terminal
$ bun run ladder show

BEGIN LADDER  v:8f21c4
## foundation
- [x] R001 Project scaffold            (small)
- [~] R002 MCP server for Claude Code  (large)
- [ ] R003 CI webhook layer            (medium) — blocked by R002
END LADDER

$ bun run ladder next --prompt   # pipe straight into Claude Code
$ bun run ladder status R002 done

How it fits your loop

  1. 01

    Ask for the next rung

    Claude Code calls ladder_next over MCP — or you run the CLI. It gets full context, the why, and the options you rejected.

  2. 02

    Build it

    The rung arrives as a ready-made prompt. Nothing to re-explain, no scrollback archaeology.

  3. 03

    Ledger updates itself

    Status and notes flow back from the terminal, and merged PRs close their own rungs via webhooks.

MCP + CLI

Native tools for Claude Code and Codex: show, next, list, status, note, add — with a verbatim render contract so the ladder is never summarised away.

Live sync

Update a rung from the terminal and the board moves in the same second. No refresh, no drift between window and shell.

CI webhooks

Point GitHub or your CI at a signed endpoint. Merge a PR that mentions R012 and the rung marks itself done, with an automation note attached.

Rungs

One decision per rung: context, why, the options considered, and which path won.

Dependencies

Blocked-by links and spinoff branches, rendered as a tree you can actually read.

Auto-Run (beta)

Opt-in mode that lets AI attempt unblocked rungs and logs every attempt.

Start with one rung

Record the decision you are about to make, then let your AI pair pull it from the terminal.

Sign in