Docs over SSH for Agents
SSH interface for docs matches how agents naturally explore codebases with grep and cat.
inplan — a Markdown editor for human ⇄ coding-agent planning. CLI + desktop editor.
Line-anchored comments on plan documents solve AI agent context drift permanently.
Teams using AI coding agents, developers frustrated with linear chat workflows
Claude Code · Cursor · Notion
We kept hitting the same failure mode with coding agents: you hand over a vague plan, the agent fills the gaps with guesses, and the back-and-forth that explains why a decision was made lives in a linear chat you can't navigate later. So we built inplan: a Markdown editor where you and the agent edit one plan document together.
The part we care most about: you can select any sentence and open a threaded comment on it — ask a question, get the agent's take, make the call — and that exchange stays anchored to that line as a durable record. When the agent changes the plan it comes back as a diff you review and apply, instead of silently rewriting the doc. It installs as a Claude Code skill (also Codex/Pi), so a "plan X" prompt opens the editor: the agent drafts, you refine inline, it revises.
It's open-core (AGPL-3.0) and runs fully locally — npm i -g inplan. There's a hosted edition (inplan.ai) for live multi-person collaboration, which is how we intend to fund it; nothing in the demo needs an account.
Where it's still young: it's an Electron desktop app, the diff is line-based, and it's built around one plan doc at a time — multi-doc and richer agent loops are next. The Windows install path was rough until last week.
60-second demo: https://github.com/melly-lgtm/inplan/releases/download/v0.1.18/inplan-demo.mp4
Would genuinely love feedback — especially where the workflow gets in your way.
SSH interface for docs matches how agents naturally explore codebases with grep and cat.
Finally, inline comments on AI plans before the agent wastes tokens writing bad code.
Discovery plans, Delivery executes — separation of concerns baked into folder structure.
Auto-synced infra-as-Markdown for Claude Code reduces agent context discovery time.
Treats Markdown files as a database so fresh agents don't need to read 70k token chat logs.
Collaborative markdown editing when Notion and GitHub wikis already exist.