Thanks for your interest in improving naive-user. Issues and pull requests are welcome.
- Be respectful. This project follows the Code of Conduct.
- For security issues, do not open a public issue. See SECURITY.md.
- Keep changes focused. One concern per pull request is easier to review and ship.
Core content lives once, and each harness gets a thin manifest that points at it. When you change behavior, you usually edit one of these two files:
skills/naive-user/SKILL.mdis the source-blind testing methodology.commands/naive-test.md(pluscommands/naive-test.tomlfor Codex and OpenCode) is the/naive-testentry point.
The harness manifests (.claude-plugin/, .codex-plugin/, .github/plugin/,
gemini-extension.json, opencode.json) only point at those files and declare the Playwright
MCP server. They should not duplicate methodology content.
- Fork and branch from
main. - Make your edit. If you change the methodology, keep it source-blind and config-driven, the two properties the whole project depends on.
- If you touch the Playwright MCP wiring, update it in every harness manifest that declares it so the harnesses stay in sync.
- Update CHANGELOG.md under
## [Unreleased]. If you use Claude Code, the/update-changelogcommand in.claude/commands/does this for you from your staged diff. - Open a pull request using the template. Describe what a user will notice and which harnesses you tested against.
There is no build step. To test end to end, point the plugin at a real running web app with a
naive-user.config.json and run /naive-test. Confirm the agent stays source-blind, signs in
via the configured auth steps, and writes a findings report under qa/naive-user/<app>/.
- Plain, direct prose. No em dashes in docs.
- Match the voice of the surrounding file.
- Markdown for docs, valid JSON or TOML for manifests.