菜单
Structured debugging workflow — reproduce first, gather evidence, trace root cause, fix with minimal scope. Use when the user asks to debug, investigate, trace, or fix a bug.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Diagnose and fix failed SkyTest test cases. Analyzes run reports, events, screenshots, and error traces to classify failures into a systematic taxonomy — then proposes targeted fixes and applies them via MCP after user confirmation. Default stabilization strategy: convert flaky ai-action interaction steps (clicks, fills, dropdowns, datepickers) to playwright-code with verified selectors. Use when a test run has failed and you need to investigate and fix the test case.
Create, update, run, and manage SkyTest test cases via MCP — directly from the user's instructions, with no separate explore/plan phase. Accepts short prompts ("add a test that creates a product and verifies it in the list") or exact step lists; fills in element detail from existing test cases, screenshots, or a quick browser-tool pass. Writes playwright-code steps for deterministic interactions (clicks, fills, dropdowns, datepickers) and ai-action for visual verification. Also handles section coverage design, project configs, runner inventory, run monitoring, and stop operations. For diagnosing a failed run, use skytest-fix.
Plan logical commit units from staged changes only, suggest commit titles, return the plan only, and create commits only after user approval (e.g. "ok", "please commit"). Use when the user asks to commit changes, plan/split commits, create logical commits, or use a refs footer (e.g. refs: #42).
Create an implementation plan for a task — align on intent, identify impacted files, and write a step-by-step plan. Use when the user asks to plan work, design a feature, create an implementation plan, or tackle a non-trivial change.
Structured code review with two-pass approach (spec compliance then code quality). Use when the user asks to review code, review a PR, audit changes, check for security/performance/reliability issues, or run a code quality check.
Convert a Linear bug report into one or more SkyTest regression test cases — verifying the bug is fixed and the expected behavior works correctly. Use this skill whenever the user shares a Linear issue link or ID that is a BUG and wants to create a regression test from it. Triggers on phrases like "create a test from this bug", "regression test for SKY-123", "convert this bug to SkyTest", "make a test case for this bug ticket", or when a Linear bug link is shared alongside testing intent. This skill is designed for bug reports specifically — not for feature tasks, user stories, or development requirements. Supports create/update, direct run trigger, run querying, project config management, and runner inventory lookup through SkyTest MCP tools.
| name | debug |
| description | Structured debugging workflow — reproduce first, gather evidence, trace root cause, fix with minimal scope. Use when the user asks to debug, investigate, trace, or fix a bug. |
Reproduce the issue, gather evidence, trace the root cause, then apply the smallest fix. No guessing, no unrelated changes.
Before changing code, capture:
Use the repo's purpose-built diagnostics first:
| Symptom | Tool / doc |
|---|---|
npm run verify failure | npm run --workspace @skytest/web verify:explain — attributes the failure to its checker |
| Runs stuck queued / claiming / dispatch issues | docs/maintainers/runner-queue-diagnostics.md |
| Frontend ↔ runtime integration (SSE, run status, artifacts) | docs/maintainers/frontend-runtime-debugging.md |
| Android device/emulator runtime | docs/maintainers/android-runtime-maintenance.md |
| Unsure where a behavior lives | Task Routing table in CLAUDE.md |
git stash push -u -m "wip backup" or a WIP commitgit restore, git checkout -- <file>, git reset --hard, git clean, git rebase, force-push) without explicit user permission