CodexQA v0.3.0

Eight published skills under codexqa-* names, including architecture wiki. Docs: https://openqa.cn/

What you can do now

• Install the router (it can fetch a worker on demand):

npx skills add openqa-cn/codexqa --skill codexqa-skill-router

• Use the new names: --skill codexqa-defect-analyzer, codexqa-code-reviewer, codexqa-code-wiki, codexqa-code-analyzer, codexqa-rootcause-analyzer, codexqa-requirement-analyzer, codexqa-testcase-generator, codexqa-testdata-generator.
• Build a DeepWiki-style architecture wiki with codexqa-code-wiki.
• Run Agent LLM Detection / judgment on the host model (no API key).

Docs

• Product site: https://openqa.cn/
• GitHub source: https://github.com/openqa-cn/codexqa
• Changelog: https://github.com/openqa-cn/codexqa/blob/main/CHANGELOG.md#030--2026-09-21

Hosts and limits

• Skills target Cursor, Claude Code, Codex, and OpenClaw. This tag does not add a new multi-host accuracy benchmark.
• Findings stay candidates for human review. README sample HTML reports are illustrations, not recorded Agent scores.
• Live @openqa-cn/codexqa indexing is not run by this repository’s CI.

Highlights

• README is a landing page. Root README leads with who it is for, eight skill HTML report screenshots, a three-step start, and activity badges. Documentation still points at openqa.cn.
• Search landing is openqa.cn. README Documentation points at openqa.cn. There is no second docs host in this repository.
• New codexqa-code-wiki skill. Index a local repository, export Leiden communities and digests with wiki inputs (no model), and write a DeepWiki-style HTML architecture wiki (sidebar + article + TOC, module map, real deps, reading guides).
• Skill rename to codexqa-*. Every published skill directory and frontmatter name now uses the codexqa- prefix. Update install flags (--skill codexqa-defect-analyzer, etc.).
• codexqa-code-reviewer Agent LLM judgment (v0.0.3). Host embedded model runs an order-16 semantic CR pass; no external LLM API required for collect/validate/render.
• codexqa-defect-analyzer Agent LLM Detection (v0.0.2). First-class detection dimension; default --llm-mode agent needs no API key.
• codexqa-testcase-generator rewritten to Plan / Exec / Incremental. Local Markdown plus aggregated HTML case report. No case-platform binding.
• New codexqa-code-analyzer skill. Local symbol-graph review of blast radius, regression scope, and test gaps. Indexing runs without an LLM.
• Repository renamed from openqa-skills to codexqa. Install path is npx skills add openqa-cn/codexqa.