| name | run-test-process |
| description | Orchestrate the full Japanese QA workflow by invoking existing skills in order, from test plan creation through reviews, analysis, design, test cases, code-based and Playwright E2E test implementation, execution recording, final HTML report generation, and final report review. Use when Codex needs to run the whole test artifact lifecycle with provided specifications and test targets while leaving human-executed tests unexecuted and visible in the final report. Runs as a single user-visible workflow while using checkpointed phase handoffs to keep long sessions reliable. |
Run Test Workflow
Overview
Run the complete test workflow by using the existing specialized skills in a fixed order. This skill is an orchestrator and does not replace the detailed instructions in each child skill; load each child SKILL.md before executing that step and follow its input, output, review-loop, and stop-condition rules.
The user-facing experience should be one-shot: when the user asks for $run-test-process, continue through the final report whenever meaningful progress is possible. Internally, treat the workflow as serial phase workers with checkpointed handoffs, so later steps use saved artifacts instead of relying on a long conversation history.
Required Inputs
Before starting, confirm the user has provided or the repository already contains:
- Product specifications, README, or equivalent reference documents.
- Test target code, URL, HTML, app entrypoint, component, or page.
- Codebase test implementation target.
- E2E automated test implementation target.
If a target needed for implementation is missing and cannot be safely inferred, do not invent it. Record the affected cases as unimplemented or unexecuted and continue toward the final report when possible.
Workflow Order
Run these steps in this exact order:
$create-test-plan$review-test-plan$create-test-analysis$review-test-analysis$create-test-design$review-test-design$create-test-cases$review-test-cases$review-test-artifacts- Implementation Entry Gate (use
$review-test-artifacts gate output; do not implement tests until it passes)
$create-test-code$review-test-code$execute-codebase-tests$create-playwright-e2e-tests$review-playwright-e2e-tests$execute-playwright-e2e-tests$create-test-report$review-test-report
Checkpointed Phase Execution
Run the workflow in these phases. Complete each phase, update the checkpoint files, then start the next phase from the saved artifacts and checkpoint state:
- Plan:
$create-test-plan + $review-test-plan
- Analysis:
$create-test-analysis + $review-test-analysis
- Design:
$create-test-design + $review-test-design
- Cases and implementation gate:
$create-test-cases + $review-test-cases + $review-test-artifacts + Implementation Entry Gate
- Codebase automation:
$create-test-code + $review-test-code + $execute-codebase-tests
- Playwright E2E automation:
$create-playwright-e2e-tests + $review-playwright-e2e-tests + $execute-playwright-e2e-tests
- Final report:
$create-test-report + $review-test-report
At the end of every phase, create or update:
ใในใๆๆ็ฉ/run-test-process_้ฒ่ก็ถๆณ.mdใในใๆๆ็ฉ/run-test-process_ๅผใ็ถใ.md
The progress file should record:
- Completed phase and timestamp.
- Generated or updated artifact paths.
- Review status, especially whether any fix-worthy
P0, P1, or P2 findings remain, and the Implementation Entry Gate result when applicable.
- Question-wait, unimplemented, unexecuted,
N/A, infrastructure issue, and blocker summaries.
- The exact input artifacts for the next phase.
The handoff file should be concise and task-local. It should contain only the facts the next phase needs: purpose, target files or URLs, constraints, relevant traceability IDs, unresolved decisions, latest execution folders, and the next phase's completion criteria.
When the environment supports child threads, subagents, or background tasks, run each phase in a fresh serial child context by default. Before launching the first child, read references/serial_subagents.md and follow its prompt, completion, verification, and close/archive protocol. Start exactly one child at a time; wait for it to complete; verify local artifacts, review stop conditions, gate results, and checkpoint files; close or archive that child when supported; then start the next child. When child execution tools are unavailable, continue in the same thread but explicitly treat ใในใๆๆ็ฉ/run-test-process_้ฒ่ก็ถๆณ.md, ใในใๆๆ็ฉ/run-test-process_ๅผใ็ถใ.md, and the generated QA artifacts as the source of truth for the next phase.
For each step:
- Read the child skill's
SKILL.md just before using it.
- Use the child skill's default output locations unless the user specified alternatives.
- Preserve traceability IDs across artifacts:
TAxxx, TVxxx, TDxxx, TC-*, BUG-*, and E2E-BUG-*.
- Report blockers and generated artifact paths before moving to a substantially different phase.
- Prefer saved artifacts and checkpoint files over conversational memory when deciding what the next step should do.
Gates And Continuation Rules
- After each creation step, confirm the expected artifact exists before starting the corresponding review step.
- After each phase, confirm the progress and handoff files exist before starting the next phase.
- If a phase used a child thread or subagent, do not start the next phase until that child has completed and has been closed or archived when the tool supports it.
- Review steps must repeat review/fix/re-review until each child review skill's stop condition is met. For review skills that define
P0, P1, and P2 as fix-worthy, continue until no fix-worthy P0, P1, or P2 findings remain.
- If a review uncovers a clear upstream gap, allow the relevant child review skill to update upstream QA artifacts as defined by that skill.
- If a question-wait or
่ฆ็ขบ่ช item appears, do not guess. Keep it in the questionnaire, question-wait test cases, or unimplemented test case list.
- Before
$create-test-code, run the Implementation Entry Gate:
- Confirm
Source Structure Inventory, Expected Case Yield, and Case Expansion Ledger exist.
- Cross-check
Source Structure Inventory -> Expected Case Yield -> Case Expansion Ledger by SSIxxx/TDxxx.
- Confirm every
TDxxx has ๆๅพ
TCๆฐ, ๅฎTCๆฐ, and ไธ่ถณๆฐ.
- If
ๅฎTCๆฐ < ๆๅพ
TCๆฐ without a concrete representative extraction reason, aggregate judgment rule, or linked ่ณชๅๅพ
ใก, return to $create-test-design, $review-test-design, $create-test-cases, and $review-test-cases as needed.
- If high-risk source structures are compressed without rationale/residual risk, return upstream before implementation.
- Do not accept "one or more
TC-* per TDxxx" as sufficient unless the expected-vs-actual ledger also passes.
- If a codebase or E2E implementation target is missing, skip only that implementation/execution lane, update the unimplemented or unexecuted records, and continue to
$create-test-report.
- If dependencies, Playwright browsers, or execution environments are missing, do not install them without user approval. Let the execution-recording skills capture
N/A or infrastructure issues when appropriate.
- Do not execute human-run test cases. Keep
ใในใใฑใผใน_ไบบ้ๅฎ่ก.md as a created and reviewed artifact, and let the final report show those cases as not yet executed.
Phase Details
Planning Through Test Cases
$create-test-plan creates ใในใๆๆ็ฉ/ใในใ่จ็ปๆธ.md.$review-test-plan saves ใในใๆๆ็ฉ/ใในใ่จ็ปใฌใใฅใผ็ตๆ.md.$create-test-analysis creates ใในใๆๆ็ฉ/ใในใๅๆ.md and ใในใๆๆ็ฉ/ใในใๅๆ_่ณชๅ็ฅจ.md.$review-test-analysis saves ใในใๆๆ็ฉ/ใในใๅๆใฌใใฅใผ็ตๆ.md.$create-test-design creates ใในใๆๆ็ฉ/ใในใ่จญ่จ.md and ใในใๆๆ็ฉ/ใในใ่จญ่จ_่ณชๅ็ฅจ.md.$review-test-design saves ใในใๆๆ็ฉ/ใในใ่จญ่จใฌใใฅใผ็ตๆ.md.$create-test-cases creates split test case files for codebase, E2E automated, human execution, and question-wait cases.$review-test-cases saves ใในใๆๆ็ฉ/ใในใใฑใผในใฌใใฅใผ็ตๆ.md.$review-test-artifacts reviews plan, analysis, design, and test cases together, then saves ใในใๆๆ็ฉ/ใในใๆๆ็ฉๆจชๆญใฌใใฅใผ็ตๆ.md.- The Implementation Entry Gate must explicitly state whether implementation may start. It must surface total expected cases, total actual cases, shortages, question-wait counts, and accepted representative/aggregate rationales.
Implementation And Execution
$create-test-code implements codebase automated tests for implementable ใในใใฑใผใน_ใณใผใใใผใน.md cases and records unimplemented cases.$review-test-code reviews and fixes P0/P1/P2 fix-worthy issues in the codebase test implementation and saves ใในใๆๆ็ฉ/ใในใใณใผใใฌใใฅใผ็ตๆ.md.$execute-codebase-tests runs or records codebase tests, producing a timestamped ใในใๆๆ็ฉ/report/yyyyMMddHHmmss_ใณใผใใใผในใในใ/ folder and ใณใผใใใผใน_็บ่ฆissueไธ่ฆง.md.$create-playwright-e2e-tests implements Playwright tests for implementable ใในใใฑใผใน_E2E่ชๅ.md cases and records unimplemented E2E cases.$review-playwright-e2e-tests reviews and fixes P0/P1/P2 fix-worthy issues in the Playwright E2E implementation and saves ใในใๆๆ็ฉ/Playwright_E2Eใในใใฌใใฅใผ็ตๆ.md.$execute-playwright-e2e-tests runs or records E2E tests, producing a timestamped ใในใๆๆ็ฉ/report/yyyyMMddHHmmss_e2e่ชๅใในใ/ folder and e2e_็บ่ฆissueไธ่ฆง.md.
Final Report
$create-test-report snapshots the current QA artifacts and latest execution folders into ใในใๆๆ็ฉ/report/yyyyMMddHHmmss_report/.$review-test-report reviews and fixes P0/P1/P2 fix-worthy issues in the generated report and saves ใในใๆๆ็ฉ/ใในใใฌใใผใใฌใใฅใผ็ตๆ.md.- The final report should make remaining
Fail, N/A, not-yet-executed, question-wait, and unimplemented cases visible.
- Human-executed tests remain unexecuted in this workflow and should appear as not-yet-executed unless the user separately provides execution results.
Final Outputs
At the end, summarize the key artifact paths:
ใในใๆๆ็ฉ/run-test-process_้ฒ่ก็ถๆณ.md and ใในใๆๆ็ฉ/run-test-process_ๅผใ็ถใ.md.- Test plan, analysis, design, and test case Markdown files.
- Review result Markdown files.
- Codebase and E2E test implementation result files.
- Unimplemented test case lists.
- Codebase and E2E execution result report folders.
- Final HTML test execution report folder and
index.html.
- Final test report review result Markdown file.
Safety
- Do not edit specifications, README files, or product code unless the user explicitly asks.
- Do not install dependencies or browsers without explicit approval.
- Do not treat skipped human-executed tests as failures; show them as not yet executed in the report.
- Continue as far as meaningful even when some lanes are incomplete, so the final report captures the current state instead of hiding gaps.
