built by aktagon.com

A regex cannot tell you whether the writing is any good.

aigrd scores each file against a rubric you wrote from your own best work — a pass or fail per criterion, with a quoted span as evidence and a note on how to fix it.

aigrd is a quality gate for agent output, judged by a second model. Below the bar, the Claude Code Stop hook blocks the turn and the agent redoes the work. aigrd judges once per call. The harness owns the loop.

Judge the work, not the diff. · One call. The harness owns the loop.

MIT 26s and $0.027 per article on haiku

01 For work only a reader can grade

A rubric is a contract the judge can check.

  • Five to ten criteria, each one decidable about the file alone.
  • Each marked must or should, with a hint saying where to look.
  • A failed must is an error; failed shoulds count against a threshold.

The judge sees the file and the criteria and nothing else, so a verdict cites the span it judged.

02 Measure the judge, not just the work

Turn "the judge seems harsh" into a number.

Every judged file appends one line to .aigrd/runs.jsonl — model, thinking cap, cost, duration, verdict. Label the verdicts you disagreed with and aigrd agreement scores the judge against you, overall and per criterion.

0  above the bar1  criterion failed2  config error

03 Every capability

One binary. From rubric to hook.

init

Write aigrd.toml, docs/rubrics/ and the .gitignore line.

rubric derive

Draft a rubric from your best pieces, then edit it by hand.

rubric check

Warn on any criterion a regex could have decided instead.

judge

Score every file that matches a genre. Verdicts cached on content.

hooks claude

Print the Stop-hook entry to paste. It never writes settings.

agreement

Compare your labels with the judge's, per criterion and per model.

Gate the prose like you gate the build.

MIT licensed. The judge runs on your existing Claude login. Not yet publicly released.