菜单
Invoke explicitly with `$bug-flow` for unclear bugs, regressions, or failure investigations that need structured diagnosis, regression tests, the smallest fix, and findings-first review. Do not use for obvious one-line fixes or planned feature work.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Use when acting as `acceptance_test_writer` to propose happy-path acceptance and integration coverage from an approved spec or ExecPlan. Do not use to materialize tests directly or to edit production code.
Use when acting as `edge_case_test_writer` to propose adversarial, boundary, and invariant coverage that complements acceptance coverage. Do not use to materialize tests or edit production code.
Invoke explicitly with `$feature-flow` for complex features, significant refactors, or multi-file behavior changes that need an ExecPlan, proposed test coverage, RED-phase failing tests, implementation, and review. Do not use for tiny fixes, doc-only work, or purely local cleanup.
Use when acting as `spec_writer` to create or refresh the active ExecPlan, clarify scope, and record acceptance criteria for a multi-agent ticket. Do not use for production-code edits or standalone prose cleanup.
Use Browser MCP for exploratory browser debugging, live-session inspection, parity compare, design review, and selector or flow discovery before converting stable paths into Playwright.
Use when acting as `browser_debugger` to gather browser evidence, run the governed QA passes, and report PASS/FAIL/BLOCKED outcomes for UI-facing work. Do not use for application-code edits.
基于 SOC 职业分类
| name | bug-flow |
| description | Invoke explicitly with `$bug-flow` for unclear bugs, regressions, or failure investigations that need structured diagnosis, regression tests, the smallest fix, and findings-first review. Do not use for obvious one-line fixes or planned feature work. |
Use this skill explicitly when the failure mode is not already fully understood and the work needs a diagnosis-first workflow.
/hyperopen/AGENTS.md/hyperopen/docs/MULTI_AGENT.md/hyperopen/docs/PLANS.md/hyperopen/docs/WORK_TRACKING.mddebugger to isolate likely root cause from code, logs, and local evidence.browser_debugger in parallel. Wait for both before deciding on the fix path.tdd_test_writer for a deterministic regression test when the behavior can be exercised automatically. If the bug is not a good automated-test candidate, record that explicitly and carry the manual validation plan forward.worker for the smallest fix that satisfies the regression contract.reviewer after the fix. Re-run browser_debugger when browser evidence is needed for final validation.name values.