App Explorer

Overview

This skill explores a web application through Chrome MCP browser automation using a breadth-first strategy. It builds a structured inventory of pages, interactive elements, and navigation edges without performing risky actions. The output is written in a stable schema for downstream skills.

Prerequisites

• Chrome MCP access is required. Read references/chrome-mcp-compat.md to detect and adapt to the available variant.
• You need the base URL to start exploration.
• If authentication is required, ask the user for credentials before continuing.

Reference Loading

• Read references/chrome-mcp-compat.md.
• Read references/exploration-strategy.md.
• Read references/multi-role-protocol.md.
• Read references/edge-cases.md.
• Read references/dangerous-actions.md.
• Read references/output-format.md.

Quick Start

1. Detect the available Chrome MCP variant using references/chrome-mcp-compat.md.
2. Navigate to the base URL and capture the initial snapshot.
3. If login is detected, ask the user for credentials and authenticate.
4. Start the BFS exploration loop with guardrails enabled.
5. Save results to .claude/artifacts/exploration-{timestamp}/.

Exploration Workflow

Step 1: Initialize

• Detect Chrome MCP at runtime.
• Navigate to the base URL.
• Capture the first snapshot and initialize tracking structures.

Step 2: Extract Elements

• Parse snapshot output and catalog interactive elements.
• Record for each element: uid, role, name, type, interactable, attributes, text_content.
• Limit capture to maxElements per page.

Step 3: BFS Queue

• Add unvisited navigation targets to the BFS queue.
• Track visited URLs in a Set.
• Normalize URLs before comparison to avoid duplicate traversal.

Step 4: Navigate and Explore

• Pop the next URL from the queue.
• Navigate, snapshot, extract elements, and append newly discovered links.
• Record navigation edges from source page to destination page.

Step 5: Repeat

• Continue until queue is empty or a guardrail limit is reached.
• Stop early on timeout or repeated navigation failure.

Step 6: Save Output

• Build the final exploration object.
• Validate structure against references/output-format.md.
• Write artifacts to .claude/artifacts/exploration-{timestamp}/.

Guardrails

• maxDepth: 3 (default BFS depth from base URL)
• maxPages: 50 (default total pages)
• maxElements: 200 (default per page)
• timeout: 30 minutes
• CRITICAL: Read references/dangerous-actions.md and enforce it strictly.
• Never click buttons with text containing: delete, remove, submit, purchase, send, confirm, pay, cancel, destroy, drop.
• Never submit forms without explicit user approval.

Multi-Role Exploration

• If multiple roles are requested, run one full exploration per role.
• Use isolation guidance from references/multi-role-protocol.md.
• Keep sessions independent and outputs separate.
• Save one output file per role under the same timestamped artifact directory.

Edge Cases

Read references/edge-cases.md and apply its handling for SPAs, hash routing, modals, infinite scroll, shadow DOM, iframes, auth expiry, dynamic content, dialogs, and tab changes.

Output

Save exploration results to .claude/artifacts/exploration-{timestamp}/ and follow the schema in references/output-format.md.

Troubleshooting

• No Chrome MCP detected: Ask the user to install a supported Chrome MCP.
• Login screen detected: Ask the user for credentials.
• Exploration timeout: Increase timeout or reduce maxDepth.
• Too many pages: Reduce maxPages or target specific sections.