app-controls
Inspect and drive Poracode itself — the Terminal panel, other threads, projects, workspaces, git, pull requests, and schedules —…
Use the user's real Chrome tabs and signed-in sessions for visible browser workflows. Use when existing authentication, open tabs, cookies, or extensions matter; prefer a purpose-built connector for semantic service operations unless the user explicitly asks for Chrome.
$ npx -y skills add Porabuild/Poracode --skill chrome-control --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/chrome-controlContext preview
The summary Claude sees to decide when to auto-load this skill.
Use the user's real Chrome tabs and signed-in sessions for visible browser workflows. Use when existing authentication, open tabs, cookies, or extensions matter; prefer a purpose-built connector for semantic service operations unless the user explicitly asks for Chrome.
name: chrome-control description: Use the user's real Chrome tabs and signed-in sessions for visible browser workflows. Use when existing authentication, open tabs, cookies, or extensions matter; prefer a purpose-built connector for semantic service operations unless the user explicitly asks for Chrome.
Use Poracode's `chrome` MCP when a task depends on the user's current Chrome tabs, authenticated sessions, or installed extensions. If a purpose-built connector can complete a semantic service operation, prefer it unless the user explicitly requested Chrome or visual interaction is part of the task.
1. Call `chrome.status` first. If the extension is disconnected, ask the user to connect it rather than switching surfaces silently. 2. Call `chrome.enable` once before browser actions. Use the background Poracode workspace by default; call `chrome.attach` only when the user asked to operate an existing tab. 3. Inspect with `chrome.snapshot` or `chrome.find` before clicking or typing. Prefer returned element refs and use `fill` for replacement versus `type` for appending. 4. Page commands share the in-app browser names and arguments: `snapshot`, `find`, `fill`, `type`, `click`, `press`, `wait`, `perform`. Batch known actions with `{steps:[{action:"fill",ref:"@e1",text:"Ada"},{action:"click",ref:"@e2"},{action:"wait",text:"Saved"}]}`. `perform` stops on failure and returns one compact snapshot (`observe:"none"` omits it). Split at decisions, new targets, and navigation; never replay completed steps after partial failure. Keep every action scoped to the requested site and task. Do not explore other tabs or signed-in content for extra context. 5. Include a condition `wait` at the end of a batch for asynchronous changes and verify the returned observation. After navigation, wait for and verify the resulting URL, text, control state, or screenshot. 6. Call `chrome.disable` before asking the user for input, waiting on an external event, or finishing.
Report the target site or tab, what changed, and the final visible evidence. State whether the background workspace or an existing user tab was used, and identify any action left pending for confirmation.
One window for all your AI coding agents. Run Claude, Codex, OpenCode, Gemini, Antigravity, Cursor, and Copilot side-by-side. Terminal and chat, any layout.
Inspect and drive Poracode itself — the Terminal panel, other threads, projects, workspaces, git, pull requests, and schedules —…
Create a deliberately staged commit from the current worktree with a clear subject and why-focused body. Use when asked to commit…
Commit and push the current worktree and create or update a pull request with an accurate title, description, verification…
Finalize the current worktree through deep review, manual QA, fixes, checks, commit, push, PR creation or update, and merge after…
Open, inspect, interact with, and verify websites or local web apps in Poracode's isolated browser. Use for visible page state,…
Prove a web change actually works by exercising the flow in Poracode's browser and collecting visual, console, and network…