git:20260920.21f92dc to git:20261010.d4b822e

1 added, 1 removed. Audit A to A.

# qa-skills

- > QA skills for AI coding agents: 12 skills + a shared knowledge base (`core/`) that make any Agent Skills host — Claude Code, Cursor, Codex, OpenCode and 70+ more — work like a senior software testing engineer. Full lifecycle: requirement analysis with clarification checkpoints, risk-based test strategy with a 10-type decision matrix, executable test cases (markmap for humans + schema.yaml for machines), case review, Playwright/pytest automation, bug analysis down to the failing line, regression scoping, flaky/suite-reliability governance, reporting, plus `qa-memory`, a cross-session project QA knowledge base. Free, MIT, zero runtime, zero telemetry. Measured On/Off gains (pre-registered gates, unfavorable numbers disclosed): case-spec compliance 0.26 → 0.98; planted-bug detection 75%; test-type decision recall 0 → 0.88 on the weakest model (first-round measurement, not yet in the formal gain table); coverage +8.7 to +13.2pp; honest cost: ~3.3x tokens. Primary language is Chinese; an English layer is on the roadmap.

+ > QA skills for AI coding agents: 12 skills + a shared knowledge base (`core/`) that make any Agent Skills host — Claude Code, Cursor, Codex, OpenCode and 70+ more — work like a senior software testing engineer. Full lifecycle: requirement analysis with clarification checkpoints, risk-based test strategy with a 10-type decision matrix, executable test cases (markmap for humans + schema.yaml for machines), case review, Playwright/pytest automation, bug analysis down to the failing line, regression scoping, flaky/suite-reliability governance, reporting, plus `qa-memory`, a cross-session project QA knowledge base. Free, MIT, zero runtime, zero telemetry. Measured On/Off gains (pre-registered gates, unfavorable numbers disclosed): case-spec compliance 0.26 → 0.98; planted-bug detection 75%; test-type decision recall 0 → 0.88 on the weakest model (first-round measurement, not yet in the formal gain table); coverage +8.7 to +13.2pp; honest cost: ~3.3x tokens. Primary language is Chinese; the landing page and README are available in English, 11/12 skill descriptions are bilingual, and instruction bodies remain Chinese.

## Install

```bash

npx skills add fishzjp/qa-skills --skill '*'

```

Verified full install: all 12 skills + `core/` land together; `core/` must always accompany any single-skill install or relative-path references break.

## Key links

- [GitHub repository](https://github.com/fishzjp/qa-skills): source, issues, discussions

- [README（中文，含完整实测数据表）](https://github.com/fishzjp/qa-skills/blob/main/README.md): measured results with per-metric methodology notes

- [README (English)](https://github.com/fishzjp/qa-skills/blob/main/README.en.md)

- [Skill On/Off output comparison](https://github.com/fishzjp/qa-skills/tree/main/examples): same PRD, with and without the skills

- [Releases](https://github.com/fishzjp/qa-skills/releases): milestone releases attach cross-model gain-matrix snapshots

- [Landing page](https://fishzjp.github.io/qa-skills/): product overview, FAQ, install