The tool now grades your plugin. claude code's september release adds plugin eval, which runs your…
the tool now grades your plugin. claude code's september release adds plugin eval, which runs your plugin against a test suite and compares the score to a no-plugin baseline, and the init command writes the test cases for you.
dogfooding became a command.
Context
Anthropic's plugin evals docs say claude plugin eval runs a plugin against a suite of test cases and scores the results, where each case is a prompt plus graders and runs three times by default, with the runs repeated with no plugin loaded so there are two scores, with and without, and a delta. The init command asks about the plugin, proposes cases and graders, tries them and writes the files. It needs Claude Code v2.1.269 or later, and every eval run and judge grader is a real model call counted against plan usage or the API bill. The v2.1.269 GitHub release of 11 September 2026 lists claude plugin eval as added.
The command, baseline and init are first-party. Init is interactive: it asks the author what a good result looks like, proposes cases and graders and runs a trial, so the author supplies input. The eval measures whether a plugin changes Claude's scored behavior on cases the author defines, and the docs say a high score alone does not show the plugin helped, which is why the baseline arm exists. It is not free, since a suite costs about cases times runs with the plugin and the same again for the baseline, plus judge calls. Community tools for evaluating plugins existed separately and were not compared. The docs page is an undated snapshot fetched on 4 October 2026, and the only dated first-party evidence for 11 September is the v2.1.269 release line adding claude plugin eval, which does not describe the baseline or init. So eval, baseline and init exist in the docs now, but that all three shipped on 11 September as described is unverified.
Related work
- Earlier note on plugin eval ↗Same command.
- Earlier note on plugin grading ↗Same command.
Watch next
- Docs changes to run counts or grader types and release notes after v2.1.269.
Sources
- Plugin evals (Claude Code docs)code.claude.com
- Claude Code v2.1.269 release (GitHub)github.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 21 September 2026 at 17:43 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →