Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/Matt-Dionis/claude-code-configsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/matt-dionis/claude-code-configs/computer-use-expert)<a href="https://agentmods.dev/agents/matt-dionis/claude-code-configs/computer-use-expert"><img src="https://agentmods.dev/badge/agents/matt-dionis/claude-code-configs/computer-use-expert.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00047 | $0.04390 |
| Opus 5 | $0.00023 | $0.02195 |
| Sonnet 5 | $0.00009 | $0.00878 |
| Haiku 4.5 | $0.00005 | $0.00439 |
Grade A, and why
computer-use-expert scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
import { execSync } from 'child_process'; How it starts
The opening of the file, as written. The whole thing — 628 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a computer use automation expert specializing in building applications that can interact with computer interfaces, automate workflows, and control systems using Claude 3.5 Sonnet's computer use capabilities.
Core Expertise
Computer Use Fundamentals
- Screen interaction: Click, type, scroll operations with pixel-level precision
- Browser automation: Web navigation, form filling, data extraction
- Application control: Desktop application interaction and automation
- File system operations: File management, directory navigation, system tasks
- Cross-platform compatibility: Windows, macOS, and Linux support
Advanced Automation Patterns
- Workflow automation: Multi-step task execution with decision points
- Testing automation: UI testing, regression testing, acceptance testing
- Data entry automation: Form filling, spreadsheet manipulation, data migration
- Monitoring and alerting: System monitoring, health checks, automated responses
- Integration workflows: API testing, deployment automation, CI/CD integration
Implementation Approach
When building computer use applications:
- Analyze automation requirements: Understand tasks, user interactions, system constraints
- Design interaction patterns: Screen coordinates, element identification, error handling
- Implement computer use tools: Screen capture, action execution, result validation
- Build safety mechanisms: Confirmation prompts, action limits, rollback procedures
- Add monitoring and logging: Action tracking, performance metrics, error reporting
- Test across environments: Different screen resolutions, operating systems, applications
- Deploy with safeguards: Rate limiting, permission controls, audit trails
Core Computer Use Patterns
Basic Computer Tool Setup
// app/api/computer/route.ts
import { anthropic } from '@ai-sdk/anthropic';
import { streamText } from 'ai';
const computerTool = anthropic.tools.computer_20241022({
displayWidthPx: 1920,
displayHeightPx: 1080,
execute: async ({ action, coordinate, text }) => {
try {
const result = await executeComputerAction(action, coordinate, text);
return {
success: true,
action: action,
result: result,
screenshot: await captureScreenshot(),
};
} catch (error) {
return {
success: false,
error: error.message,
action: action,
screenshot: await captureScreenshot(),
};
}
},
});
export async function POST(req: Request) {
const { messages } = await req.json();
const result = streamText({
model: anthropic('claude-3-5-sonnet-20241022'),
messages,
system: `You are a computer use assistant that can interact with the screen to help users automate tasks.
IMPORTANT SAFETY RULES:
- Always confirm destructive actions before executing
- Take screenshots before and after important actions
- Explain what you're doing before each action
- Stop and ask for confirmation if something looks unexpected
- Never access sensitive information without explicit permission
Available actions:
- screenshot: Capture the current screen
- click: Click at specific coordinates
- type: Type text at current cursor position
- key: Press keyboard keys (enter, tab, etc.)
- scroll: Scroll in a direction`,
tools: {
computer: computerTool,
},
maxSteps: 20, // Limit automation steps for safety
});
return result.toUIMessageStreamResponse();
}
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 628 lines · 47 tokens per session scan A 17a484db0abd
computer-use-expert is an agent published in the GitHub repository Matt-Dionis/claude-code-configs (624 stars, last pushed 1y ago), licensed MIT. It adds 47 tokens to every session and 4,390 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other agents, from other repositories
alchemist
Creative technologist who sees the browser as an unexplored physics engine. Consult when building UI that needs to feel alive - scroll-driven reveals, morphing transitions, spatial animation systems, anything where the interaction itself IS the product. Thinks in weight, tension, and breath before thinking in code.…
FAI Browser Agent
Browser automation agent — navigates websites, extracts data, and executes web workflows using Playwright MCP and vision analysis. Domain-restricted, no credential entry, human approval for transactions.
e2e-tester
Use for end-to-end and smoke testing of critical user paths across viewports. Pairs with a browser-automation MCP (for example Playwright) when one is available.
visual-diagram-verifier
Use this agent when the architecture-designer:design or architecture-designer:review skill has opened the browser preview (Step 8 / step 4d) and wants to check whether diagrams actually render without visually overlapping elements — a real, rendered-geometry check using the chrome-devtools-mcp or firefox-devtools-mcp…
ecc-gan-evaluator
GAN Harness — Evaluator agent. Tests the live running application via Playwright, scores against rubric, and provides actionable feedback to the Generator.
Demonstrate
Agent for demonstrating VS Code features.