Tools Reference

Run shell commands with auto-detected shell. Shell detection: CMD syntax → CMD, PowerShell syntax → PS, neutral commands (npm/git/python) → PS, shell: "wsl" → Linux. Supports background mode for servers and long-running tasks.

exec("Get-Process | Sort-Object CPU -Descending | Select-Object -First 10")

Manage background processes started with exec. List running processes, poll for new output, wait for completion, kill, or write to stdin.

process({ action: "poll", session_id: "bg_1" })

Write files directly into the WSL filesystem with proper Windows-to-Linux path translation. Avoids shell-escaping issues with heredocs and echo.

write_wsl_file("/home/user/script.sh", "#!/bin/bash\necho hello", executable: true)

Read file contents with line numbers and pagination. Supports partial reads with offset/limit for large files.

read_file("C:/Users/me/project/src/main.ts", offset: 50, limit: 100)

Create or overwrite files with full content. Auto-creates parent directories. Auto-runs syntax checks on .py, .json, .yaml, .toml files.

write_file("config.json", "{\n  \"port\": 3000\n}")

Targeted edits using fuzzy matching (9 strategies). Supports find+replace, multi-edit batches, line-range mode, and replace-all. Auto-runs syntax checks after editing.

edit_file("src/app.ts", old_string: "const x = 1", new_string: "const x = 2")

Search for text within a single file. Returns matching lines with surrounding context. Use before read_file on large files to find exact line numbers.

search_in_file("src/main.ts", "handleClick")

List files and directories sorted by modification time. Supports recursive listing, glob filtering, and depth control.

list_directory("C:/Users/me/project", recursive: true, pattern: "*.ts")

Ripgrep-grade search across files in a directory tree. Supports regex patterns, glob filtering, context lines, and multiple output modes.

search_files("TODO|\bfix\b", path: "src/", file_glob: "*.ts", context: 2)

Create an empty directory. Idempotent — does not error if the directory already exists.

create_directory("C:/Users/me/project/src/components")

Create a new file with content. Fails if the file already exists — use write_file to overwrite.

create_file("src/utils.ts", "export const greet = (name: string) => \`Hello, ${name}!`")

Append text to the end of an existing file. Creates the file if it does not exist.

append_file("notes.md", "\n## New Section\nAdded content.")

Delete a file with automatic backup before removal. Accidental deletions can be recovered via restore_file.

delete_file("C:/Users/me/project/old-config.json")

Move or rename files and directories. Auto-creates destination directories.

move_file("src/old-name.ts", "src/new-name.ts")

Copy files to a new location. Auto-creates destination directories.

copy_file("config.json", "config.backup.json")

Restore a file from its most recent backup created by delete_file or edit_file.

restore_file("src/old-config.json")

List available backups for a file. Shows timestamps and sizes of each backup.

list_file_backups("src/main.ts")

Open a native Windows file picker dialog for user file selection. Supports file type filters and multi-select.

open_file_dialog(filters: [{ name: "Images", extensions: ["png", "jpg"] }])

Open a native Windows save dialog for choosing save locations and file names.

save_file_dialog(default_name: "report.pdf")

Create, list, diff, or restore filesystem checkpoints for safe rollback. Snapshots file state before risky operations.

checkpoint_manager({ action: "create", paths: ["src/", "config.json"] })

Search the web and return structured results with titles, URLs, and snippets. Use for up-to-date external information.

web_search("how to configure Tailwind CSS v4")

Fetch and extract webpage content as LLM-optimized structured text. Automatically strips ads, navigation, and clutter.

read_webpage("https://blog.example.com/article")

Fetch raw webpage content without extraction. Returns the full HTML response.

fetch_webpage("https://api.example.com/data")

Open a URL in the system default browser or a local file in its default application.

open_url("https://windop.app")

Download a file from a URL to a local path. Supports progress tracking for large files.

download_file("https://example.com/data.csv", "downloads/data.csv")

Search the local documentation library using semantic/vector search. Returns relevant pages from previously crawled sources.

search_library("React hooks useEffect cleanup")

Fetch the full content of a specific documentation page by URL from the local library.

pull_library_doc("https://react.dev/reference/react/useEffect")

List all documentation sources in the library with metadata (page count, last sync, status).

list_library_sources()

Re-crawl and re-index a documentation source to update the local library.

sync_library_source("react-docs")

Create a new documentation source in the library by providing a base URL and crawl configuration.

create_library_source({ name: "Next.js Docs", base_url: "https://nextjs.org/docs" })

End-to-end documentation crawler — crawls a site, scrapes pages, and stores them in the local library.

crawl_and_store("https://docs.example.com", max_pages: 50)

Open or navigate the controlled browser to a URL. Returns a compact page snapshot with interactive elements and ref IDs.

browser_navigate("https://example.com")

Get a text-based snapshot of the current page's accessibility tree. Returns interactive elements with ref IDs for click/type actions.

browser_snapshot(full: true)

Click an interactive element from the latest browser snapshot by its ref ID (e.g., @e5).

browser_click("@e5")

Type text into an input element from the browser snapshot by ref ID. Clears the field first, then types the new text.

browser_type("@e3", "search query text")

Press a keyboard key in the controlled browser. Useful for submitting forms (Enter), navigating (Tab), or shortcuts.

browser_press("Enter")

Take a screenshot of the current browser page for vision analysis. Returns the image for visual inspection.

browser_screenshot()

Unified desktop session tool for full-PC screen viewing and mouse/keyboard automation. Uses AI vision to understand your screen like a human would. Actions: screenshot, find_element, click, move, drag, scroll, type, key, wait, cursor_position.

computer_control("screenshot") → computer_control("find_element", query: "Settings button") → computer_control("click", x: 450, y: 320)

Launch Windows applications by path or name. Auto-detects installed apps from the Start Menu and PATH.

launch_application("notepad")

Generate images from text prompts or edit existing images via OpenRouter. Supports multiple providers, style presets, and aspect ratios. Returns the image inline in the chat.

generate_image("A cyberpunk cityscape at sunset, neon lights reflecting on wet streets")

Analyze images using vision AI. Supports single images, batches, or entire directories. Returns detailed descriptions and answers questions about the image content.

analyze_image("screenshot.png", "What UI elements are visible?")

Create a detailed, optimized image generation prompt from a basic idea. Expands vague descriptions into production-ready prompts.

describe_image_for_generation("a cat wearing a top hat")

List available image generation models with their capabilities, supported styles, and pricing.

list_image_models()

Generate a short AI video through OpenRouter. Supports text-to-video and image-to-video modes.

generate_video("A timelapse of a flower blooming in a garden")

Convert text to speech using the configured TTS provider (Edge TTS, OpenAI, or custom). Returns an audio file that plays inline in the chat.

text_to_speech("Welcome to WindOp, your AI-powered desktop assistant.")

Unified Git operations: status, log, diff, branch, add, commit, push, pull. Handles authentication, upstream tracking, and merge conflict detection.

git("status") → git("commit", message: "feat: add new feature")

Execute code in a sandboxed environment. Supports Python, JavaScript, TypeScript, and Bash. Returns stdout, stderr, and exit code.

execute_code("python", "import pandas as pd\nprint(pd.DataFrame({\"a\": [1,2,3]}).describe())")

Read, write, delete, list, and search Windows Registry keys. Full access to HKLM, HKCU, and other hives.

windows_registry({ action: "read", key: "HKEY_CURRENT_USER\\Software\\Microsoft\\Windows" })

List and manage Windows services. Start, stop, restart, and inspect service status and configuration.

windows_services({ action: "list", filter: "running" })

List, create, delete, run, enable, and disable Windows Scheduled Tasks.

windows_task_scheduler({ action: "list" })

Query, list, and clear Windows Event Logs. Filter by log name, level, source, and time range.

windows_event_log({ action: "query", log_name: "Application", level: "Error", max_entries: 20 })

Run Windows network diagnostics and Wi-Fi operations. View connections, ping hosts, manage Wi-Fi profiles, and check DNS.

windows_network({ action: "diagnostics" })

Inspect and manage Windows Updates. List available updates, check update history, and trigger installations.

windows_updates({ action: "list_available" })

Inspect Windows volumes, disk usage, cleanup candidates, and defragmentation status.

windows_disk({ action: "usage" })

Inspect Windows Defender status, start scans, view threat history, and manage real-time protection settings.

windows_defender({ action: "status" })

Detect and provision Windows development runtimes using winget. Check for Node.js, Python, Rust, Go, Java, .NET, and more.

windows_dev_environment({ action: "detect" })

Inspect and customize Windows Terminal, Oh My Posh, Starship, and Nerd Fonts. Configure profiles, themes, and prompt styling.

windows_terminal_setup({ action: "status" })

Smart unified package manager supporting winget, choco, scoop, npm, pip, cargo, and go. Auto-detects the right manager for the package.

install_packages({ manager: "winget", action: "install", package: "Git.Git" })

Bridge between Windows and WSL Linux environments. List distros, run commands, copy files between filesystems, and manage WSL instances.

wsl_bridge({ action: "run_command", command: "ls -la /home", distro: "Ubuntu" })

Create a specialist agent team with custom roles and execution mode. Supports parallel, sequential, and discussion-based execution.

create_team("security-review", members: [{ role: "Code Reviewer" }, { role: "Security Analyst" }], mode: "parallel")

Execute a previously created team on a task. The supervisor coordinates members, collects results, and synthesizes a final answer.

run_team("security-review", task: "Review src/auth.ts for security issues")

Create a group chat with specialist agents for multi-turn discussion. Agents can respond to each other, building on previous messages.

create_group_chat(agents: ["Backend Expert", "Frontend Expert", "DevOps Engineer"])

Add another specialist agent to an existing group chat for expanded perspectives.

add_agent_to_chat(id: "chat_1", agent: "Security Expert")

Run a previously created group chat on a specific topic or task.

run_group_chat(id: "chat_1", task: "Discuss the best database for our use case")

Inspect current group chats, teams, and temporary specialist agents for this session.

get_autogen_status()

Manage bounded long-term memory. Actions: add (new entry), replace (update existing), remove (delete). Categories: preference, knowledge, pattern, insight.

write_to_memory("preference", "User prefers concise responses with code examples")

Add timestamped entries to the daily log file. Entry types: event, decision, learning, task, note.

create_daily_log("event", "Deployed v2.1.0 to production")

Search across all past conversations using full-text search. Returns matching messages with context from historical sessions.

session_search("Docker deployment config")

Search across all past conversations with advanced filtering. Finds historical context, decisions, and code snippets.

search_past_conversations("React useCallback optimization")

Load a skill's full instructions and supporting files. Skills provide specialized knowledge for specific task types.

load_skill("react-performance")

List all available skills (global + project-scoped). Shows name, scope, description, and trigger patterns.

list_skills()

Create a new reusable skill with markdown instructions, trigger patterns, and optional supporting files.

create_skill({ name: "api-patterns", description: "REST API conventions", content: "..." })

Search configured skill hub sources for community-shared installable skills.

search_skill_hub("Docker deployment")

Install a skill from the hub or a URL. Skills are stored locally and can be loaded on demand.

install_skill({ query: "kubernetes" })

Check for and apply updates to hub-managed skills. Can target a specific skill or check all.

update_skills({ apply: true, skillName: "docker-patterns" })

Archive a skill without permanently deleting it. The skill can be restored if needed.

delete_skill("old-api-patterns")

Create, fetch, and list local kanban boards. Boards track project tasks across customizable columns.

kanban_board({ action: "create", name: "Sprint 12", columns: ["Backlog", "In Progress", "Review", "Done"] })

Create, update, move, assign, decompose, block, unblock, complete, and delete tasks on kanban boards.

kanban_task({ action: "create", board_id: "board_1", title: "Implement auth", column: "Backlog" })

Dispatch kanban tasks to configured sub-agents for automated execution. Agents pick up tasks and report progress.

kanban_dispatch({ board_id: "board_1", task_id: "task_5", agent: "coder" })

MCP integration manager — setup, list, import, remove, connect, disconnect, and refresh MCP server connections. Supports Claude Code, Codex CLI, and local tools.

manage_mcp_servers({ action: "list" })

Manage WindOp's own MCP server mode, allowing external MCP clients (Claude Code, Codex, etc.) to connect and use WindOp's tools.

manage_windop_mcp_server({ action: "status" })

Manage local webhook endpoints — create, list, delete, and test webhook listeners for event-driven automation.

manage_webhooks({ action: "list" })

Get local usage metrics including token consumption, estimated costs, tool usage frequency, and latency statistics.

get_metrics({ range: "today" })

Get a detailed local trace for a conversation turn — shows tool calls, latencies, token usage, and decision flow.

get_trace({ conversation_id: "conv_123" })

Pause execution and ask the user a question. Returns their free-form response. Use when you need clarification, a decision, or credentials.

ask_human("Which database should we use: PostgreSQL or SQLite?")