Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchAI agents in Visual Studio Code use a language model, project context, and tools to pursue a development goal. Instead of stopping after suggesting code, an agent can inspect files, search the project, edit code, run commands, examine the results, and decide whether another step is needed. The exact tools, models, controls, and customization options depend on the selected agent harness, your account, and your organization’s policies.
What an AI agent does in VS Code
Visual Studio Code describes an agent as an AI system that uses a language model and tools to complete a goal on your behalf. The key difference from a simple text response is the ability to take actions in an environment and use the results of those actions to guide further work.
For example, you might ask: “Find the cause of the failing tests in this project, fix it, and run the relevant tests to verify the change.” An agent can inspect relevant files, search for related code, propose or make a change, run tests if an appropriate tool is enabled, and interpret the output. It may repeat that cycle, ask for confirmation, or stop if it cannot make progress.
This is not a guarantee of a correct fix or a fully autonomous process. The agent can misunderstand the goal, choose an unhelpful file or command, or interpret output incorrectly. Treat its actions and results as work to review, not as proof that the project is correct.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How the agent tool loop works
- You state a goal. Describe the result you want, the scope, and any constraints. “Fix the tests” is less precise than identifying the failing test, asking for the smallest relevant fix, and specifying which test command to run.
- The agent gathers context. It may use enabled tools to search code, read files, inspect editor information, or obtain other project context.
- It chooses an action. The model decides which available tool, if any, could help. Depending on enabled tools, an action might read or edit a file, run a terminal command, or query an external service.
- The tool returns a result. The agent receives output such as matching code, a changed file, test results, or a service response.
- The agent evaluates and continues or stops. It can use the result to select another action, revise its approach, ask you a question, or report what it found.
The loop is useful because actions can supply fresh evidence: a test failure may reveal the next file to inspect, while a successful run may help establish that a change worked for that particular test command. It does not establish that every relevant test ran or that no other defect remains. Review what ran and what changed.
Which tools can an agent use?
VS Code documents three broad sources of tools. The enabled set is what matters in a particular session; not every harness or installation exposes the same choices.
| Tool source | What it can provide | What to check |
|---|---|---|
| Built-in tools | Common development actions such as file operations, terminal use, code search, and editor navigation. | Which actions are available to the selected agent and which require approval. |
| MCP tools | Tools supplied by Model Context Protocol servers, which can connect an agent to data or external services. | Which server and tools are enabled, what information they can access, and whether calls require confirmation. |
| Extension-contributed tools | Tools contributed by VS Code extensions through the Language Model Tools API. | What the extension exposes and what permissions or external connections its tools involve. |
An agent generally selects from the tools enabled for it. You can also direct it to a tool with a # tool reference where the selected experience supports that syntax. A tool being enabled means it is available; it does not mean every call is automatically approved. Availability and approval are separate settings.
How to give an agent a useful task
Good requests define an outcome and boundaries without prescribing guesses about the cause. Include relevant files, commands, and constraints when you know them; ask the agent to inspect rather than assume when you do not.
Recommended Free Tools
Example: investigate a failing test
Find the cause of the failing tests in this project. Inspect the relevant code and test output, make the smallest appropriate fix, and run the relevant tests. Do not change unrelated files. Before using commands that modify data or access external services, ask me first. Report the files changed and the exact tests run.The request explicitly narrows the change and asks for an auditable report. It cannot enforce approval behavior by itself: configure the harness’s actual approval controls for sensitive actions rather than relying on prompt wording.
Make the result verifiable
- Name the expected outcome, not only the activity: for example, “the targeted test passes” rather than “look at the tests.”
- Set scope: identify a package, directory, issue, or file when appropriate, and say when unrelated changes are out of bounds.
- Ask the agent to report commands and files changed so you can check its account against the editor and terminal state.
- For an uncertain diagnosis, request findings before edits, or split investigation and implementation into separate tasks.
- Use tool references when you need to steer toward a particular enabled tool; do not assume a reference can enable a tool that is unavailable.
Reusable roles with custom agents
A custom agent packages a recurring role—such as planning, implementation, or code review—along with its instructions and an optional selection of tools. VS Code’s documented format is a Markdown file that can include an optional YAML header for metadata and tool selection. This lets you reuse role-specific guidance rather than restating it for every task.
For relevant agent-host sessions, the VS Code documentation lists .github/agents as a workspace location and ~/.copilot/agents or ~/.claude/agents as user locations. These paths and the controls around custom agents depend on the selected harness. Check the current VS Code documentation and the selected provider’s guidance before relying on an exact location or format.
What to put in a role definition
- Role and scope: state the kind of work the agent should do and where it should work.
- Working rules: explain constraints such as preserving public interfaces, avoiding unrelated edits, or reporting uncertainty.
- Tool selection: select only tools appropriate to the role where the harness supports tool selection.
- Review expectations: ask for a concise account of files changed, commands run, and unresolved issues.
A role file guides behavior; it is not a security boundary. Use actual permissions, approval prompts, and sandboxing to control access.
Free tools Windows power users keep installed
One-click scans. No signup required.
Approvals and sandboxing are different controls
VS Code documents both approval prompts and sandboxing. They address different questions: approval concerns whether an action needs your confirmation; sandboxing limits the resources a terminal command can access. One does not replace the other.
Approval prompts
Approval settings can require you to review tool names and input parameters before calls, particularly for edits, terminal commands, and external services. Read what the proposed action will do before approving it. A request for approval is a chance to inspect an action, not evidence that the action is harmless.
Sandboxing
Sandboxing restricts filesystem and network resources accessible to terminal commands, including after a command has been approved. It can limit the reach of a command, but does not tell you whether the command is appropriate or whether its output is correct. Keep review in place as well.
A practical control checklist
- Enable only tools needed for the task, especially when an agent can reach external services or sensitive project data.
- Use confirmation for consequential edits, terminal commands, and service calls when the harness offers that control.
- Use sandbox restrictions for terminal filesystem and network access where available; check what the selected harness actually enforces.
- Inspect diffs and generated files before integrating the work.
- Review command output and independently verify the relevant tests or behavior before treating a task as complete.
Harnesses, accounts, and where work runs
VS Code names Copilot, Claude, and Codex among the available agent options, but their availability and capabilities are not necessarily identical. VS Code states that available models, tools, and customizations depend on the selected harness, your account, and your organization’s policies. An organization can therefore affect what an individual user sees or can enable.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsWhen assessing an agent option, check these factors in your actual VS Code session and provider account:
- Whether the harness is available to you and permitted by your organization.
- Which built-in, MCP, and extension-contributed tools it exposes, and how you enable or manage them.
- Where the language model is hosted and where tool execution takes place. These are separate questions: a hosted model does not, by itself, establish where every tool runs or what data it can access.
- Which custom-agent formats, handoffs, and permission controls the harness supports.
Do not infer identical feature support or privacy behavior from a product name alone. Verify current provider and VS Code settings for the harness and account you intend to use.
Adding website screenshots to an agent workflow
If a development task needs a rendered website image—for example, to inspect a page layout—an agent needs an enabled tool that can capture or retrieve that image. ScreenshotNeo is a website screenshot API and MCP server for developers, made by Yorker Media. Its MCP server provides take_screenshot, get_page_info, and capture_pdf tools for AI agents, including Claude, Cursor, and any MCP client. Whether it is usable from your particular VS Code agent depends on that harness’s MCP support, account, and organization policy. See ScreenshotNeo and its documentation for setup details; no MCP configuration is assumed here.
For a direct API call outside the agent tool loop, this cURL request saves a screenshot of Stripe’s homepage as WebP:
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webpReplace YOUR_API_KEY with your key. The API also accepts the parameter names used by other screenshot APIs, which can make switching easier. Its response includes X-Page-Verdict and X-Billed headers; consult the documentation for the request and response details.
Or skip the browser setup
ScreenshotNeo accepts a URL in one GET request and returns a PNG, JPEG, WebP, or PDF. Before capture, it accepts cookie or consent banners like a visitor and removes more than 60 known consent platforms, newsletter popups, and chat widgets; each cleanup step can be turned off. Bot checks or CAPTCHAs, blank pages, timeouts, failed loads, and cache hits cost nothing, and the response identifies the page verdict and billing status in headers. An MCP server gives AI agents screenshot, page-info, and PDF tools. The free plan includes 1,000 screenshots per month with no card; paid plans start at $5 for 3,000 screenshots. Get 1,000 free screenshots a month with no card.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Troubleshooting an agent workflow
The agent cannot find or use a tool
Check that the selected harness supports the tool source, that the tool or MCP server is configured and enabled, and that the account or organization permits it. A # reference can direct an agent to an available tool; it cannot make an unavailable tool appear.
The agent asks for approval or stops before acting
Inspect the pending tool name and parameters, then approve only if the requested action fits the task. If a call is blocked or unavailable, check the harness’s approval controls and policy rather than weakening access broadly.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →A command fails inside the sandbox
Determine which filesystem or network resource the command needs and whether sandbox restrictions block it. Adjust access only as narrowly as the task permits. Approval alone does not remove sandbox limits.
The agent says a fix worked, but the result is uncertain
Check the diff, the exact command and output, and whether the test actually covers the behavior in question. Ask the agent to investigate a specific failure or run a specified relevant test; do not treat a confident summary as a substitute for verification.
A custom agent is not recognized
Confirm that the file uses the format and location supported by the current harness. The documented workspace and user-level paths vary by agent-host session, so a location valid for one harness may not apply to another.
Using agents reliably in a development workflow
Agents are most useful when a task has inspectable inputs, a bounded scope, and a way to check the outcome. Start with a narrow request, let the agent gather evidence, and inspect proposed changes before integrating them. For broad or risky work, divide the job into discovery, implementation, and verification rather than granting a single vague goal wide access.
Reliability also depends on the tools available and the evidence they return. A test command can verify only the tests it ran; a screenshot can show a rendered state but not prove that every interaction works. Tell the agent what evidence matters, and check that evidence yourself when the impact warrants it. Model hosting and tool execution location should be considered separately when deciding whether a workflow is appropriate for your code or data.
Frequently Asked Questions
Does an AI agent in VS Code always make changes automatically?
No. What it can do depends on enabled tools and harness controls; edits and other actions may require confirmation, and some tools may be unavailable.
Can a custom agent definition enforce file or network security?
No. Instructions and tool selection shape a role, but access restrictions must come from the harness’s permission and sandbox controls.
Does using an agent prove that the code is correct?
No. Review its changes and activity, and verify the relevant behavior with appropriate checks.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
