wimi321

Use when the user wants a large, mechanical change split into many independent units and executed in parallel with isolated workers and PRs.

Use when the user wants browser automation in an existing Chrome session: click, fill, inspect tabs, capture screenshots, or debug web flows.

Use when code imports Anthropic SDKs or the user asks to build with the Claude API, Agent SDK, tool use, streaming, files, or batches.

Use when onboarding a repository into Claude Code by creating or improving CLAUDE.md, and optionally identifying useful skills, hooks, and verifier setup.

Use when the user wants to update Claude settings, hooks, permissions, MCP server toggles, or other JSON config safely and with scope awareness.

Use when changed code should be reviewed for reuse, code quality, and efficiency, then cleaned up before finalizing.

Use when the user wants the full git workflow: branch creation if needed, commit, push, and PR create or update with a concise summary and test plan.

Use when the user wants a single clean commit created from current changes with safe staging, message drafting, and non-interactive git usage.

Top-level macOS computer-use skill with a bundled standalone runtime that bootstraps itself without any local Claude installation, private native modules, or extracted app assets.

Test Capacitor, Ionic, and WebView hybrid apps on real Android devices using Chrome DevTools Protocol (CDP) via ADB. Use when the user wants to test a mobile app on a physical device, debug WebView content, interact with React or Vue state on a real phone, take device screenshots, or test offline and network scenarios. Handles the core challenge that uiautomator cannot see WebView content by using CDP for web elements and uiautomator for native system dialogs.

Use when a successful session or repeatable process should be captured as a reusable skill with steps, arguments, and invocation guidance.

Use when a project needs one or more verifier skills for web, CLI, or API runtime checks that the Verify agent can discover automatically.

Use when another Claude Code session appears frozen, stuck, or abnormally slow and needs process-level diagnosis and reporting.

Use when the user wants to configure Claude Code statusline UI by inspecting shell prompt configuration and updating the relevant settings.

Use when diagnosing a failing or confusing Claude Code session by enabling debug logging, tailing recent logs, and explaining warnings or errors.

Use when the current branch or PR needs a focused security review that minimizes false positives and only reports concrete, exploit-relevant issues.

Use when a code change must be verified by actually running the app, endpoint, or CLI flow instead of relying only on unit tests.

Use when the user wants to create, inspect, update, or run scheduled remote agents in the cloud rather than local cron jobs.

Use when the user wants a prompt or command to run on a recurring interval, such as checking deploys, polling status, or repeating a slash-command.