Understand. Review. Improve.

AI builds it.
You stay in control.

Claude wrote your app. Pen Wiki shows what it built, and what changed since the last time you looked.

  1. 1UnderstandDocuments

    One document explains your project: what it does, its features, how it fits together.

  2. 2ReviewGuides · Runs

    Run a guide again and again. Each Run says what changed since the last one: fixed, new, worse.

  3. 3ImproveFix with Claude

    Point at a finding and press Fix with Claude. It shows its plan, waits for your OK, and works on a branch of its own.

For people who build with Claude Code, and the teams that review their work. Nothing runs until you press Run.

Pen Wiki

Runs in VS Code with your own Claude Code. Nothing leaves your machine until you publish a report.

Install in VS Code Free

Pen Wiki for Teams

Publish a report, ask a teammate to review it in the browser, and keep the history of every Run.

Register $8 / month
Pen Wiki in VS Code on a laptop: a sentence selected, a comment typed on it, and the kinds of answer to choose from
Adam, founder of Pen Wiki
I vibe-coded my apps.
But I wanted to understand them more.
Adam, Founder of Pen Wiki

Run it again.
See what changed.

How it works, on one example: a review you run before every release.

1
Pen Wiki

Run a guide

Press Run on Review My Project. Your Claude reads the code and writes a report: ten checks, each with its result and the file it points at. What it could not verify, it says so.

2
A week later

Run it again

The new report starts with what changed since the last one, not the whole report again. No git needed: every report remembers the run before it.

Changes since Run 1Run 2 compared with Run 1: 1 fixed · 1 new · 1 worse · 7 still openFixed: C2 Every route checks who asksNew: C11 Rate limits on the login routeWorse: C5 Errors show no internals (pass → fail)
3
Pen Wiki for Teams

Ask for a second opinion

Publish the report and ask a teammate to review it. They read and comment in the browser, nothing to install. Their comments come back to your editor and wait for your OK.

Guides

Your rules. Your Claude. Repeatable results.

A prompt is asked once. A guide is your team's recipe: a Markdown page your Claude Code runs again and again as the project grows. Write it once, press Run, review what comes back.

Guide

The same report, every time.

Write down once what a good report looks like: the checks, the sections, the format. Your Claude reads your code and writes it. Run it again and the new report starts with what changed.

review-my-project.md run #2 · 1 fixed · 1 new · 1 worse
  • Your team's standard, not whatever the AI felt like today.
  • No git needed: every report remembers the run before it.
  • Publish a report and ask a teammate to review it in the browser.

Good for architecture · security reviews · standards · onboarding · release notes

Fix with Claude

Fix what the report found.

Point at a finding and press Fix with Claude. It shows its plan and waits for your OK, works on a branch of its own, and tells you which files changed and which checks ran.

C5 · Errors show internals 2 files · tests passed
  • Claude shows its plan and waits for your OK.
  • It works on a branch of its own and never pushes.
  • Each check is passed, failed or not run, never guessed.

Good for fixes · missing tests · small refactors

The document drives the next step.

  1. GuideReview

    Claude checks the code against your rules and writes the review.

  2. YouDecide

    Comment on the findings. Pick what gets fixed.

  3. FixFix

    Press Fix with Claude on a finding: your OK first, then its own branch and the checks.

  4. GuideCompare

    Run the review again: what is fixed, what is new, what is still open.

Where it pays off

WorkA guide reportsFix with Claude
Your vibe codeAn architecture tour and a map of what depends on whatSimplifies the parts nobody understands
A new featureThe spec and its acceptance criteriaBuilds it, with its tests
SecurityA security review: what passes, what does not, whereFixes the findings you picked
Your standardsWhere the project departs from your rulesBrings the code in line
BugsThe root cause, written downThe fix, and a test that keeps it fixed
TestsWhere coverage is thinWrites the missing tests

More than a prompt file: every run leaves a report your team comments on and reviews, and it says what changed since the run before. The guide itself is reviewed like any page, so the rules are agreed before Claude follows them.