ReadMe Realist
Runs in the cloud on every PR — or locally on every commit

Never Ship Stale Docs.

Keep your documentation in sync with your codebase. Run it fully managed in the cloud via GitHub App, or execute locally inside Claude Code with zero subscription costs.

Pull Request #482 · feature/redis-cache
readme-realist commented
Drift detected

This pull request adds a new environment variable, but README.md never mentions it.

+ REDIS_CACHE_URL = "redis://localhost:6379"
readme-realist / drift-check · neutral, awaiting update

What ReadMe Realist posts on your pull request

Built on tools your team already uses

GitHubDockerPythonFastAPIGoogle GeminiAnthropic

The problem, and the fix

Code ships every day. The docs that explain it don’t. That gap is what ReadMe Realist closes.

The problem: docs drift

Pull requests add settings, flags, and routes every day. Updating the README is the step everyone forgets.

  • New settings get added to the code but never to the README.
  • New teammates hit setup steps that no longer work.
  • Code review catches bugs, not missing docs.

The fix: ReadMe Realist

A GitHub check that reads each pull request, works out what changed, and compares it to your docs using an AI model.

  • Pulls the settings, flags, and routes out of the code change.
  • Checks them against your docs and writes the missing lines for you.
  • Ignores formatting-only and docs-only changes, so it stays cheap.

First, it reads what changed

It skips lockfiles and binary files, then scans the rest for the things that change how your project runs — settings, flags, routes, and dependencies.

Comment edits, renamed variables, and whitespace are dropped before the AI model ever sees the change.

See it in the diagram

Found in app/config.py

env: REDIS_CACHE_URLenv: MAX_RETRIESflag: --enable-metricsroute: /healthz
Change + Docs
AI model
Verdict

Then it checks the docs

The changes, plus your current docs, go to Gemini or Claude in a single request.

The model replies in a fixed format, not loose text, so the check can act on it without guesswork.

Try it in the simulator

How it works

From the moment GitHub pings us to the comment on your pull request — every step, in the order it runs.

Show this as a plain diagram instead
Rendering diagram…

Try it yourself

Pick a change, run it, and watch it turn into a comment on the pull request.

Pick a change
Sample
Current content
What the check is doingReady
// Ready. Pick a change below, then run it.
What it would postPR comment
Nothing run yet.

Step by step

Every pull request goes through the same three stages on its way to a posted verdict.

Receive

Steps 01–03

  1. 01GitHub sends the pull request, signed.
  2. 02We check the signature before trusting it.
  3. 03The request is cleaned up and queued.

Analyze

Steps 04–07

  1. 04A check appears on the PR; the change is downloaded.
  2. 05The change is scanned; noise is filtered out.
  3. 06Matching docs are found in the repo.
  4. 07The change and the docs go to the AI model.

Publish

Steps 08–09

  1. 08The model's answer is read into a clear verdict.
  2. 09One comment is posted or updated; the check finishes.

Two ways to run it.

Run it free on your own machine inside Claude Code, or let the managed GitHub App handle every pull request for your team.

Local CLI Skill

Free forever
$0pay only your own API tokens

Run directly on your machine using Claude Code. The skill reads your latest commit and proposes doc edits — nothing leaves your laptop.

  • 100% private — code never leaves your machine
  • Zero subscription overhead
  • Triggered locally via Git post-commit hooks
  • Unified diff preview with commit-amend support

Install the skill

mkdir -p .claude/skills/readme-updater
curl -fsSL https://readme2.kamipresents.com/skills/readme-updater/SKILL.md -o .claude/skills/readme-updater/SKILL.md
Download SKILL.md

Cloud Bot / GitHub App

Managed
$9/ mo · or custom for teams

Zero setup for team repositories. Installs directly to GitHub organizations and runs on every pull request — no key, no workflow file, no hosting.

  • Automatic checks on PR open & push webhooks
  • Zero local config across the whole engineering team
  • Server-side rate limit & queue management
  • Automated bot PR comments and branch updates
  • No LLM API key required — runs on our managed keys
Sign in to subscribe

Scopes: contents (read), pull requests (write), checks (write). Nothing else.

Frequently Asked Questions

The questions a check that reads your pull requests should answer up front.

Does my code leave my repo?

At review time, the pull request change and your documentation files are sent to the Gemini or Claude API. Nothing else is sent, and nothing is sent at any other time.

Do you store my code?

No. ReadMe Realist keeps no copy of your change or your docs once the review finishes. There is no database of your content and no content in the logs.

Can it block a merge?

No. It posts a comment and a neutral check. If the model times out or hits a rate limit, it fails open and never holds up your merge.

What does it cost?

Two options. The local Claude Code skill is free — you run it on your machine and pay only your own model tokens. The managed GitHub App is a paid plan ($9/mo, or custom for teams) and needs no API key of your own; it runs on our managed keys. Formatting-only and docs-only changes skip the model completely either way.

What permissions does it need?

The GitHub App requests read access to pull request contents and write access to checks and comments — contents: read, pull-requests: write, checks: write. Nothing else. The local skill needs no permissions; it only reads your working tree and Git history.

Will it spam my pull request?

No. It keeps one comment per pull request and edits that comment in place on every new push.

Ready to catch drift before it ships.

Install the managed GitHub App, or drop the free skill into Claude Code — either way your next change gets checked automatically.