1.0.0-beta.1 — public beta

The first tagged release. The manifests had carried a 1.0.0 since 2026-08-23 with no tag and no release behind it, so there was nothing anyone could pin or cite. This is that point.

It is a beta, and the two reasons are worth stating plainly.

What is verified

The skill is developed and tested against Claude Code. One canonical definition in skills/github-readme-generator/ serves every host, and npx skills add pekral/github-readme-generator installs it into Codex, Cursor and the other agents the skills CLI knows — but no scenario has been run there. Those hosts are packaged, not verified, and the documentation now says so wherever it used to claim otherwise.

What is not

The benchmark in tests/evals/ exists to answer whether the skill actually reduces invented claims. It does not answer that yet. Its scorer reads help text and command output as invented commands, and a repository's own logo as an invalid badge (#11) — enough to move a run's totals by more than the two modes differ. Until that is fixed, no number from it is published as evidence, here or anywhere else.

The three before/after pairs in docs/examples/ are still worth reading, and they report their own inconvenient findings: on those repositories the baseline agent invented nothing, and in one pair the skill did worse than the baseline by discarding hand-written prose its own rules tell it to keep.

Installing

npx skills add pekral/github-readme-generator

Or as a Claude Code plugin:

/plugin marketplace add pekral/github-readme-generator
/plugin install github-readme-generator@pekral

What is in it

Evidence-first README generation in three modes — create, update, audit — with per-ecosystem rules for PHP, Node, Python, Rust and Go, badge templates that emit only what the repository earns, and a public-surface audit that reports findings as text without changing a single repository setting.

Full detail in the changelog.