Clidoc

The Problem

Building with AI in the terminal is fast. Maybe too fast. You get into a flow, ship something real, close the session — and the thinking evaporates. The code is there. The reasoning isn't.

I noticed this while taking the ShipWithAI course. Session after session, I'd build something worth documenting and walk away with nothing but a git diff. No record of why I structured it that way, what I tried first, or what broke along the way. Writing it up after the fact meant reconstructing decisions I'd already moved on from.

The Idea

Git already solves half the problem — it captures what changed. What it doesn't capture is why. The decision, the tradeoff, the failed attempt before the one that worked.

Clidoc fills that gap. Run it after any terminal AI session and it reads your git diff, sends the context to an LLM, and generates a structured doc of what you built and the reasoning behind it. One command. One API key. Works with OpenAI or Anthropic.

How It Works

npx clidoc -n "what you worked on"

It reads the diff from your last commit, bundles that with your session description, and asks the LLM to produce a structured document: what was built, the decisions made, problems that came up, and what's next. Plain markdown — ready to commit, paste into a case study, or just keep as a log.

The constraint was intentional. No dashboard, no account, no app to open. It lives where the work happens — in the terminal, right after you ship.

Why It Matters

The shift to AI-assisted development is compressing how fast things get built. That speed has a cost: the reasoning layer — the part that explains why a codebase is shaped the way it is — stops getting written down. Clidoc is a small fix for that. A memory layer for the way you work, not just what you shipped.