メニュー
Use when a bug investigation requires confirming behaviour in a live browser before classifying the root cause. Produces a Behaviour Report that the Bug Investigation Specialist attaches to the ledger before classification.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | browser-reproduction |
| description | Use when a bug investigation requires confirming behaviour in a live browser before classifying the root cause. Produces a Behaviour Report that the Bug Investigation Specialist attaches to the ledger before classification. |
| argument-hint | Bug title or description to reproduce |
| user-invocable | true |
Use this skill when a bug investigation requires confirming behaviour in a live browser before classifying the root cause. Produces a Behaviour Report that the Bug Investigation Specialist attaches to the ledger before classification.
Requires the Playwright MCP server. Confirm it is running before executing any tool call below.
npm run dev (frontend on http://localhost:5173)npm run start-backend (port 3001).vscode/mcp.json)Use the Playwright MCP browser_navigate tool to open the relevant URL.
browser_navigate({ url: "http://localhost:5173/<route>" })
Record the route in the Behaviour Report.
Use browser_snapshot to capture the accessibility tree at the starting state. This provides structured text that can be parsed without screenshot OCR noise.
browser_snapshot()
Identify any visible error states, disabled controls, or missing elements noted in the bug report.
Use browser_click, browser_type, and browser_select_option to replicate the steps from the bug report exactly.
browser_click({ element: "<aria label or role description>" })
browser_type({ element: "<aria label>", text: "<input value>" })
After each user action that should trigger a change, call browser_snapshot again to capture the updated tree.
Use browser_console_messages to gather all console output accumulated since navigation.
browser_console_messages()
Filter for error and warn level entries. Record exact error text.
If the bug involves API calls, use browser_network_requests to list requests and responses during the reproduced session.
browser_network_requests()
Look for 4xx/5xx responses, missing auth headers, or CORS errors.
Use browser_take_screenshot to produce a screenshot at the failure state.
browser_take_screenshot()
Include this in the Behaviour Report as visual reference.
After completing Steps 1–6, produce a Behaviour Report in this format and attach it to the .github/.ai_ledger.md under Verification Failures:
### Behaviour Report — <Bug Title>
**Reproduction URL:** http://localhost:5173/<route>
**Steps Performed:**
1. ...
2. ...
**Observed Behaviour:**
- (What actually happened)
**Expected Behaviour:**
- (What the bug report says should happen)
**Console Errors:**
- [error] <exact error text>
- [warn] <exact warning text>
**Network Failures:**
- POST /api/<endpoint> → 401 Unauthorized
**Screenshot:** [attached above]
**Reproducible:** Yes / No / Intermittent
After the Behaviour Report, output a one-sentence Root Cause Hint for the Bug Investigation Specialist:
"Console error indicates
<component>dispatches<action>before auth state is initialized — investigate initialization order in<file>."
This hint is written to the ledger Root Cause Hint field.
Use when gathering files, running searches, and producing a structured context manifest for other agents. Read-only — never writes code or makes design decisions.
Use when solution design, architecture-fit analysis, task decomposition, or high-ambiguity tradeoff analysis needs stronger reasoning before coding starts.
Use when executing a feature end to end through plan, implement, test, docs, review, ledger updates, and bounded recursive repair.
Use when implementing code changes in a repository where domain-specific specialists are not installed. Generic — no stack assumptions. Stack context loaded from the project's .instructions.md at runtime.
Use when reviewing changes for regressions, correctness, security, and missing coverage. Generic — no stack assumptions. Stack context loaded from the project's .instructions.md at runtime.
Use when adding or repairing tests — unit, integration, or component. Generic — no stack assumptions. Stack context loaded from the project's .instructions.md at runtime.