Grade your codebase
like a report card.
Letter grades, honest scores, and a prioritized fix list for your UI, backend, code, and UX.
$ /grade everything
acme-dashboard
overall grade
- ui 88 [B+]
- backend 90 [A-]
- code 93 [A]
- ux 84 [B]
# top fix
Settings page buttons use inconsistent label casing. Normalize to sentence case.
One command. Five lenses.
Grade a single dimension or the whole project. Add a modifier to fix issues, or loop until the score stops climbing.
$ /grade ui
grade the rendered frontend, open the report
$ /grade code revise
grade, then fix the issues in one pass
$ /grade everything loop 3
grade, fix, regrade, repeat up to 3 rounds
ui
Screenshots the running app at three viewports and grades the pixels, not the JSX.
backend
Routes, models, auth, and error handling, reviewed across the server tree.
code
Structure, naming, duplication, file size, and test coverage across the repo.
ux
Drives the real flows with Playwright. Clicks, load times, and feedback.
everything
GPARuns all four and rolls them into one report with an overall grade point average.
From repo to report card
Detect
Inspects the stack and project shape, so a landing page is not judged like a full-stack app.
Grade
Scores each dimension against a detailed rubric, so grades stay consistent run to run.
Report
Writes a styled HTML report card with strengths, weaknesses, and a ranked fix list.
Drive the score up, honestly.
The loop runs grade, fix, then regrades with a fresh agent that has no memory of prior rounds. Improvements are measured, not graded on a curve. It stops when the score plateaus.
- + Fresh subagent per round removes self-grading bias
- + Stops on plateau, or after a round cap you set
- + Flags regressions if a fix makes things worse
# score progression
r1 · 72
r2 · 87
r3 · 91
$ plateau reached, stopping at round 3
Drop it in and grade.
Clone into your Claude Code skills folder, start a session, and run a grade command in any project.
git clone https://github.com/josephg29/gradeskill.git ~/.claude/skills/codebase-grader