browser

Use this agent when a Playwright flow is failing and needs automated diagnosis and repair. Diagnoses root causes using agent-browser snapshots and the interaction log, edits flow code, and re-runs until flows pass. Also use when the explore skill's flow or record mode produces a failing flow file. Triggers: "heal tests", "fix playwright", "tests are failing", "debug e2e", "playwright failures", "flaky tests", "test healing", "flow is failing". <example> Context: User runs a generated flow and it fails. user: "The flow I just generated is failing on the dashboard page." assistant: "Let me engage the playwright-healer agent to diagnose and fix the flow." </example> <example> Context: User recorded a flow but selectors broke after a code change. user: "My recording flow broke after the Angular update." assistant: "I'll use the playwright-healer agent to update the selectors." </example>

Use this agent when the explore skill enters Record mode and needs to generate a Retina-quality recording of browser interactions. Reads the interaction log, generates a Playwright `.flow.ts` with CDP screencast, runs the flow, stitches frames into MP4/GIF via ffmpeg, and returns the output path. If the flow fails, invokes the playwright-healer agent internally. Triggers: "record flow", "generate recording", "screencast", "PR evidence video", "capture browser interaction". <example> Context: The explore skill has finished inspecting and the user wants a recording. user: "Record the site creation flow." assistant: "I'll dispatch the recorder agent to generate the Retina recording." <commentary> Record mode is token-intensive (CDP frames, ffmpeg output, potential healing iterations). Dispatching to the recorder agent keeps the main context lean — only the result path comes back. </commentary> </example> <example> Context: The user wants a GIF alongside the MP4 for a PR description. user: "Record this and also make a GIF for the PR." assistant: "Dispatching the recorder agent with --gif to produce both MP4 and GIF." <commentary> Same reason as above — the GIF pipeline adds palette generation and denoise steps that would further bloat the main context if run inline. </commentary> </example> <example> Context: The user is exploring a page and switches to record mode mid-session. user: "Actually, let's record what I just did." assistant: "The interaction log has your steps. I'll dispatch the recorder agent to turn them into a video." <commentary> The interaction log already captured every agent-browser command via the PostToolUse hook. The recorder agent reads it and translates to a recording flow — no need to repeat the interactions. </commentary> </example>

Use when the user wants to inspect a web page, check UI state, verify element presence, explore an app interactively, create a replayable Playwright flow, or record a Retina-quality video of a browser interaction. Uses agent-browser for lightweight exploration (93% less context than Playwright MCP) and Playwright connectOverCDP for flows and recording — both on the same shared Chrome session. Framework-agnostic — works with any web application on any URL (localhost or remote). Triggers: "explore page", "check the UI", "what does the page look like", "snapshot", "quick look", "inspect the app", "agent-browser", "create a flow", "make this replayable", "record this", "make a video", "screen record", "capture a screencast", "PR evidence video".

Use when the user needs to authenticate into a web app using agent-browser and 1Password, open a persistent browser session, log in as a specific role, switch between users, or manage named agent-browser sessions. Handles automated login via snapshot+refs discovery, session reuse, parallel multi-role login, and already-authenticated detection. Handles form-based login including Auth0 and Angular Material via agent-browser's fill command. Triggers: "login", "browser session", "authenticate", "agent-browser", "persistent session", "op run", "log in as", "switch user", "open app", "browser login".

1 hook across 1 event