# Opentest Workflow

> Standardize OpenTest QA artifact generation, review gates, bilingual outputs, and output folder organization. Use when Codex works on OpenTest-style testing workflows, test proposals, requirement analysis, test points, test cases, test reports, Claude slash commands for OpenTest, or any task involving OpenTest artifact structure and quality gates.

- Skill: `zhou-yun-long/opentest-workflow` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add zhou-yun-long/opentest-workflow`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zhou-yun-long/opentest-workflow/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: zhou-yun-long (https://skillmd.com/u/zhou-yun-long)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/zhou-yun-long/opentest-workflow

---


# OpenTest Workflow

## Core Rules

- Treat OpenTest as a staged QA artifact pipeline. Generate artifacts directly into their stage folders.
- Do not dump new artifacts into `output/` root and do not depend on a later cleanup step.
- Keep helper scripts under `scripts/`, not under `output/`.
- Use Claude slash commands under `.claude/commands/` as the executable prompt specs for each stage.
- Use repository `docs/` only for human-facing documentation; do not load diagrams or long docs unless the user asks for explanation.

## Required Output Layout

```text
output/
  00_requirement/
  01_proposal/
  02_analysis/
  03_test_points/
  04_cases/
  05_mindmap/
  06_report/
  99_archive/
  _legacy_tools/
```

## Bilingual Policy

For every major Markdown artifact, create aligned versions:

```text
*.ai.md       English, strict, AI-oriented source for downstream commands
*.zh-CN.md    Chinese, review-oriented source for humans
```

Keep IDs aligned across both versions, including workflow IDs, testable item IDs, test case IDs, dataset IDs, risk IDs, and pending question IDs.

## Quality Gates

Do not generate executable cases until proposal and analysis are case-ready:

```text
Case Generation Ready: Yes
Readiness Score: >= 90
```

The proposal/analysis must include tested object boundary, requirement evidence, workflow plan, P0/P1/P2 priorities, data format boundary, validation oracle, test data matrix, abnormal flows, coverage gaps, pending questions, and readiness score.

## Command Flow

```text
/opentest-doctor
/opentest-run-all <requirement file or folder>
/opentest-prepare-requirement <requirement file or folder>
/opentest-new output/00_requirement/requirement_text.ai.md
/opentest-revise-proposal output/01_proposal/test_proposal.ai.md
/opentest-analysis output/01_proposal/test_proposal.ai.md
/opentest-points output/02_analysis/requirement_analysis.ai.md
/opentest-generate-cases output/02_analysis/requirement_analysis.ai.md P0
/opentest-generate-cases output/02_analysis/requirement_analysis.ai.md P1
/opentest-generate-cases output/02_analysis/requirement_analysis.ai.md P2
/opentest-localize-cases output/04_cases/test_cases.json
/opentest-validate-cases output/04_cases/test_cases.json
/opentest-generate-excel output/04_cases/test_cases.json output/04_cases/test_cases.xlsx
/opentest-generate-xmind output/04_cases/test_cases.json
/opentest-report output/04_cases/test_cases.xlsx
/opentest-archive output output/99_archive
```

Use `/opentest-run-all` when the user wants an automated end-to-end run. It must still obey the analysis gate, priority-file validation, chunked localization, and final validation rules.

## Stage Rules

- Requirement preparation normalizes PDF, OCR text, screenshots, or mixed requirement folders into `output/00_requirement/`.
- Requirement analysis must be workflow-driven, not a flat module list.
- Test points must be grouped by P0/P1/P2 workflows and must respect data format boundaries and validation oracles.
- Case generation creates stable English JSON only. Chinese display fields are added later.
- Generate cases by priority (`P0`, `P1`, `P2`) before merging. Do not create large bilingual JSON in one pass.
- `priority`, `workflow_priority`, and the `workflow_id` prefix must match (`P0-*`, `P1-*`, `P2-*`).
- Every case must have a non-empty `oracle`.
- Localize in chunks of 25 cases or fewer. If localization changes case count, IDs, English fields, priorities, or workflow IDs, discard that run and retry with a smaller chunk.
- Run deterministic validation before Excel export.

## Excel Rules

The workbook must be human-readable by default:

- `Summary`: overview counts by priority and type.
- `用例总表`: Chinese-first execution sheet with only review/execution columns.
- `AI_Source`: full bilingual/English source fields for traceability.

Do not put every English and Chinese field side by side on the primary execution sheet.

## Legacy Cleanup

Only use cleanup rules for old projects that already contain messy root-level outputs. For new work, write directly to staged folders. Do not delete locked files; ask the user to close Excel, WPS, or preview panes first.