iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Custom agents and remote delegation solve different problems in GitHub Copilot CLI. A custom agent is a named specialist profile, usually used by the local CLI as a focused subagent with its own instructions, tools, and context. /delegate, also available as &, sends a task to GitHub-hosted Copilot cloud agent, which works in the background on a branch and opens a draft pull request.
The October 28, 2025 announcement introduced both capabilities, but Copilot CLI is now generally available, as announced on February 25, 2026. The practical distinction is important: having access to Copilot CLI does not automatically guarantee access to cloud delegation, and a custom agent selected in a local CLI session should not be assumed to carry over to /delegate.
At a glance
| Capability | Where it runs | How you control it | Typical result |
|---|---|---|---|
| Custom agent in CLI | Your local Copilot CLI process or a local subagent | /agent, an explicit prompt, or --agent |
Local analysis, edits, tests, or a review |
| Local autopilot | Your local machine | Interactive mode or flags such as --autopilot |
Autonomous changes in the local working tree |
/delegate or & |
GitHub-hosted Copilot cloud agent | A delegated prompt and GitHub repository policies | A remote branch, draft pull request, and agent-session link |
| Cloud custom agent | GitHub cloud-agent environment | GitHub’s Agents interface or issue-assignment workflow | Specialized cloud-agent work, normally delivered as a pull request |
Use a custom agent when you need a repeatable specialist role. Use local autopilot when the work must happen immediately on your machine. Use /delegate when the task can continue after you close the terminal and a GitHub branch and draft pull request are useful.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11What the October 2025 announcement introduced
GitHub’s original October 28, 2025 changelog announcement added two related workflows to Copilot CLI:
#1 Best Overall
- FAST RUNS IN THE FAMILY — The 14-inch MacBook Pro with the M5 Pro or M5 Max chip brings next-generation speed and powerful on-device AI to personal, professional, and creative tasks. With all-day battery life, double the starting storage,* and a breathtaking Liquid Retina XDR display, it’s pro in every way.*
- BUCKLE UP — Along with a next-generation CPU, faster unified memory, and up to 2x faster SSD storage,* M5 Pro and M5 Max feature a more powerful GPU with a Neural Accelerator built into each core, delivering faster AI performance and on-device training capabilities. So you can blaze through demanding workloads at mind-bending speeds.
- BUILT FOR AI — Apple silicon, and every major component that powers it, is designed to run demanding on-device AI workloads like LLM inference and training. And Apple Intelligence helps you write, express yourself, and get things done effortlessly with groundbreaking privacy protections at every step.*
- ALL-DAY BATTERY LIFE — MacBook Pro delivers the same exceptional performance whether it’s running on battery or plugged in.*
- MACOS RUNS APPS FAST — All your go-to apps run lightning fast in macOS, including built-in apps like FaceTime and Messages. Plus, built-in virus protection and free software updates help keep your Mac running smoothly and securely.
- Markdown-based custom-agent profiles in user, repository, and organization scopes.
- Interactive custom-agent selection through
/agent. - Noninteractive selection through the
--agentoption. - Automatic exposure of custom agents as tools that the main agent could call when appropriate.
- The
/delegatecommand for sending work asynchronously to Copilot coding agent, now commonly described in the documentation as Copilot cloud agent. - Branch and pull-request delivery for delegated work.
- Streaming output and parallel tool calls.
That release was the launch point, not the complete current behavior. Copilot CLI entered public preview on September 25, 2025, reached general availability on February 25, 2026, and continues to change rapidly. The GitHub release page listed stable version 1.0.78, released August 3, 2026, and prerelease 1.0.79-9 on August 7, 2026, as of August 10, 2026. Do not pin an evergreen tutorial to those numbers; update the CLI or use the current installation instructions instead.
Requirements, availability, and authentication
What you need
- An active GitHub Copilot subscription.
- Linux, macOS, Windows PowerShell, or Windows Subsystem for Linux. On Windows, GitHub documents PowerShell 6 or later as a prerequisite.
- An organization or enterprise policy that allows Copilot CLI.
- For
/delegate, access to Copilot cloud agent, a paid Copilot plan or eligible organizational entitlement, repository access, and policies that permit cloud-agent use.
GitHub documents Copilot CLI as available with all Copilot plans, subject to administrator policy. That statement applies to the CLI itself, not automatically to cloud delegation. Cloud agent can be disabled at the account, organization, enterprise, or repository level, and some managed-user or explicitly restricted repository configurations are not eligible. Check GitHub’s Copilot CLI overview and cloud-agent availability documentation if the CLI works but /delegate does not.
Install or update Copilot CLI
Use whichever installation method fits the machine:
# npm; requires Node.js 22 or later
npm install -g @github/copilot
# macOS and Linux with Homebrew
brew install --cask copilot-cli
# Windows with WinGet
winget install GitHub.Copilot
# macOS and Linux installation script
curl -fsSL https://gh.io/copilot-install | bashThe installation script supports VERSION and PREFIX environment variables if you need a pinned or custom installation. For an existing installation, update with:
copilot updateThen start the CLI from a trusted repository directory:
cd /path/to/repository
copilotGitHub’s installation documentation has the current platform and authentication details. On first launch, enter /login. GitHub also documents authentication with a fine-grained personal access token containing the Copilot Requests permission, supplied through COPILOT_GITHUB_TOKEN, GH_TOKEN, or GITHUB_TOKEN. Follow the documented variable precedence rather than placing a token in a prompt or repository file.
What a custom agent is
A custom agent is more than a saved prompt or persona. It is a Markdown profile that can define a display name, required description, specialist instructions, allowed tools, an optional model, optional MCP-server configuration, and whether the main agent may invoke it automatically. Copilot generally runs custom-agent work as a subagent with its own context window. That keeps large search results, test logs, and specialist instructions from consuming the main agent’s context, and suitable subagents may run in parallel.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
That behavior is a useful default, not a promise that every request will always be handled in a separate context. The main agent can decide to answer directly, invoke a specialist, or combine several agents. Explicit selection is therefore preferable for security reviews, release checks, and other tasks where routing matters.
Where to put the profile
For Copilot CLI, the current documented locations include:
| Location | Scope |
|---|---|
.github/agents/ |
Project or repository scope; the recommended GitHub location |
.claude/agents/ |
Project scope for Claude-compatible agent files |
~/.copilot/agents/ |
User-wide local scope |
<plugin>/agents/ |
Installed-plugin scope |
The CLI command reference says it searches project agents while walking upward from the current directory to the Git root, which is useful in a monorepo. It also documents .github/agents taking precedence over .claude/agents at the same level, project agents taking precedence over user agents, and plugin agents having the lowest priority.
There is an unresolved documentation conflict: another GitHub how-to and a conceptual page describe the home-directory copy as overriding a repository copy when names collide. Do not rely on duplicate names. Use unique, lowercase, hyphenated filenames such as security-auditor.agent.md, restart the CLI after changing discovery files, and test the behavior with the CLI version you actually deploy. See the CLI command reference and custom-agent how-to for the current scope details.
Recommended profile format
Use the portable .agent.md extension. The current CLI command reference also accepts .md, but .agent.md makes the file’s purpose clear and is better suited to sharing the profile across GitHub Copilot surfaces.
Here is a deliberately restricted, read-only security-review agent:
---
name: security-auditor
description: Reviews application code for high-confidence security vulnerabilities without modifying files.
tools:
- view
- grep
- glob
infer: false
---
You are a security auditor.
Review only the files and directories named in the request.
Look for:
- Exposed secrets or credentials
- SQL injection
- Cross-site scripting
- Authentication or authorization bypasses
- Unsafe deserialization
- Command injection
- Vulnerable dependency usage
Do not modify files.
Return:
1. Finding
2. Severity
3. Evidence with file and line references
4. Exploitability
5. Recommended fix
Report only issues supported by concrete evidence.In this example, the filename is security-auditor.agent.md. The identifier used with --agent is normally the filename without .agent.md; it does not always have to match the frontmatter’s name.
According to the current references, description is required, name is optional, and omitting name causes the filename to supply the identifier and display name. If tools is omitted, the agent receives all available tools. The CLI command reference still documents infer, which defaults to enabled, while the cross-surface custom-agent configuration reference describes newer disable-model-invocation and user-invocable terminology and calls infer retired. This is a compatibility difference, not something to silently ignore: check the version-specific custom-agent configuration reference before sharing a profile across products.
Free tools Windows power users keep installed
One-click scans. No signup required.
The configuration reference allows a prompt body of up to 30,000 characters. Keep profiles focused anyway. State the scope, permitted actions, prohibited actions, expected evidence, and output format. Natural-language instructions are useful, but restricting the tools list is a stronger safety boundary than simply writing Do not edit files.
Create an agent from inside Copilot CLI
- Start Copilot CLI with
copilot. - Enter
/agent. - Choose Create new agent.
- Choose Project for
.github/agents/or User for~/.copilot/agents/. - Choose whether Copilot should generate the profile or whether you will fill it out manually.
- Specify its name, description, instructions, and tools.
- Review the resulting Markdown file and restart the CLI so it is discovered.
Generated profiles should be treated as drafts. Check whether they grant shell, edit, or network tools more broadly than intended, whether the description triggers the agent for unrelated tasks, and whether the instructions tell it to modify files when you intended a review-only role.
Invoke a custom agent
Interactive selection
Inside a Copilot CLI session, enter:
/agentSelect the agent and then enter the task. This is the most direct way to verify that the profile was discovered.
Explicit natural-language invocation
Use the security-auditor agent to inspect src/auth and report findings with file and line references.This communicates the intended specialist clearly, but the model still interprets the request. For important work, use the selector or command-line option below.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Prompt-based inference
If inference is enabled and the profile’s description matches the request, Copilot may choose the agent automatically:
Check all TypeScript files under src/ for authentication and authorization problems.Automatic selection is probabilistic. It is not a routing guarantee. Make descriptions specific about task types, directories, file types, trigger phrases, and exclusions. For a critical review, explicitly select the agent instead of hoping the model infers the right one.
Programmatic invocation
copilot --agent security-auditor --prompt 'Check src/auth for high-confidence security vulnerabilities'This is useful in scripts and repeatable engineering workflows. Remember that the process can still ask for approval when its profile permits tools that require it.
Built-in agents
The current CLI command reference lists built-in agents including:
Recommended Free Tools
explore— read-only codebase analysis.task— builds, tests, linters, formatters, and command execution.code-reviewandsecurity-review— review-oriented analysis.research— deeper repository, GitHub, and web research.general-purpose— complex, multi-step work.rubber-duck— constructive critique or a second opinion.
Names, models, and behavior are version-sensitive. Treat this list as examples from the current command reference rather than a permanent API contract. A custom agent is worthwhile when your organization needs a stable role, narrower tools, or repository-specific instructions that a built-in agent does not provide.
Delegate a task to Copilot cloud agent
Delegation moves the task out of the local CLI process. The documented interactive form is:
/delegate Add pagination to the users API. Update unit and integration tests, run the test suite, and open a draft pull request summarizing the implementation and remaining risks.The current shorthand is:
& Add pagination to the users API. Update unit and integration tests, run the test suite, and open a draft pull request summarizing the implementation and remaining risks.These are equivalent forms. A good delegated prompt is self-contained: name the intended files or component, define acceptance criteria, identify tests and commands to run, state what must not change, and ask for risks or unresolved failures in the pull-request summary.
What happens after delegation
- Copilot asks whether to commit unstaged changes as a checkpoint.
- Copilot creates a new branch.
- Copilot cloud agent works remotely and opens a draft pull request.
- The task continues in the background, so closing the local terminal or shutting down the computer does not stop the remote work.
- Copilot supplies links to the pull request and the agent session.
- You review the branch and can request further work through the pull request or agent session.
This is not simply the same local agent running in a second terminal. The execution environment, available credentials, repository policies, permissions, billing, and review path all change. A local file in ~/.copilot/agents or a local MCP configuration is machine- or user-scoped; it should not be assumed to be available in the cloud-agent environment. For cloud-specific customization, use GitHub’s documented Agents interface and issue controls, and review the cloud custom-agent documentation.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Can /delegate use the custom agent I selected locally?
Do not assume it can. Copilot CLI documents local custom-agent selection with /agent, explicit prompts, and --agent. The current delegation documentation documents a prompt after /delegate or &, but does not document a /delegate --agent NAME option.
You may write:
& Use the repository security-auditor profile to review the authentication changes, run the tests, and open a draft pull request with the findings.That may communicate intent, but it should not be described as guaranteed agent selection. If deterministic specialization is required in the cloud, select the custom agent through the GitHub cloud-agent interface or issue-assignment workflow. Keep local CLI selection and cloud-agent selection as separate controls.
Custom agent, autopilot, or delegation?
| Choose | Best when | Important trade-off |
|---|---|---|
| Custom agent | You repeatedly need a specialist such as a security auditor, API reviewer, migration planner, or documentation maintainer. | It improves consistency and can restrict tools, but model-based inference is not deterministic unless you select it explicitly. |
| Local autopilot | The task should run immediately on your machine and you want it to use the local checkout, tools, and environment. | It stops when local execution stops, and broad permissions can cause destructive local changes. |
/delegate or & |
The task is well-defined, can run remotely, and should result in a GitHub branch and draft pull request. | It depends on cloud-agent entitlement, repository policy, GitHub Actions capacity, remote setup, and AI-credit availability. |
Local autopilot can be enabled interactively by cycling modes with Shift+Tab, or from the command line:
copilot --autopilot --yolo --max-autopilot-continues 10 -p 'Run the test suite and fix only failures in the authentication package'Autopilot is appropriate only for well-defined, reversible work. The --yolo flag and similar broad-approval options increase risk; they do not make the model more accurate. Use a disposable branch or worktree and inspect the result.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsCost, limits, and permissions
AI credits and Actions minutes
Current GitHub billing documentation describes Copilot usage in terms of AI credits, not the older premium-request terminology found in legacy pages. CLI interactions consume credits according to factors such as model, input and output tokens, cached tokens, and context size. Cloud-agent sessions consume AI credits and GitHub Actions minutes, and total usage depends on the model and how long the remote task runs. Exact entitlements differ by individual, Business, and Enterprise billing arrangements.
Inside a CLI session, inspect usage with:
/usageYou can apply a soft per-response credit limit from the command line:
copilot -p 'Run the test suite and fix only failures in the authentication package' --max-ai-credits 50Or set it during a session:
/limits set max-ai-credits 50The limit is soft: an in-progress response may finish slightly above the configured amount. See GitHub’s CLI usage documentation, session-limit documentation, and cloud-agent billing documentation for current rules. Avoid copying old quota numbers into a current guide.
Tool approval and local safety
Copilot CLI can read and modify files and execute commands below the trusted directory. Start it from the repository you intend to work on, not from your home directory or a directory containing unrelated secrets.
Use the narrowest permissions possible:
- Give a review-only agent read and search tools rather than edit or shell tools.
- Use a disposable branch or worktree for agents that can modify files.
- Review shell commands, dependency changes, generated workflows, and files outside the requested scope.
- Consider local sandboxing with
/sandbox enablewhen appropriate. - Do not place credentials in prompts, agent profiles, or untrusted repository files.
- Avoid
--allow-all-tools,/allow-all, and/yolofor unfamiliar repositories. These options can remove approval prompts and permit broad command execution.
GitHub recommends trusted directories and appropriate local or cloud sandboxing in its Copilot CLI security guidance. Cloud agent may automatically run CodeQL, dependency, secret-scanning, and code-review checks, but those checks are validation aids, not a guarantee that generated code is secure. Review every resulting pull request before merging; GitHub describes the automated protections in its cloud-agent validation announcement.
Rank #3
- FAST RUNS IN THE FAMILY — The 16-inch MacBook Pro with the M5 Pro or M5 Max chip brings next-generation speed and powerful on-device AI to personal, professional, and creative tasks. With all-day battery life, double the starting storage,* and a breathtaking Liquid Retina XDR display, it’s pro in every way.*
- BUCKLE UP — Along with a next-generation CPU, faster unified memory, and up to 2x faster SSD storage,* M5 Pro and M5 Max feature a more powerful GPU with a Neural Accelerator built into each core, delivering faster AI performance and on-device training capabilities. So you can blaze through demanding workloads at mind-bending speeds.
- BUILT FOR AI — Apple silicon, and every major component that powers it, is designed to run demanding on-device AI workloads like LLM inference and training. And Apple Intelligence helps you write, express yourself, and get things done effortlessly with groundbreaking privacy protections at every step.*
- ALL-DAY BATTERY LIFE — MacBook Pro delivers the same exceptional performance whether it’s running on battery or plugged in.*
- MACOS RUNS APPS FAST — All your go-to apps run lightning fast in macOS, including built-in apps like FaceTime and Messages. Plus, built-in virus protection and free software updates help keep your Mac running smoothly and securely.
A reliable end-to-end local workflow
- Update the CLI. Run
copilot updateor use the installation method appropriate to your platform. - Open the correct repository. Start Copilot from a trusted project directory so project-level agents can be discovered.
- Create the profile. Use
/agentand choose Create new agent, or add a reviewed file such as.github/agents/security-auditor.agent.md. - Restart Copilot. This avoids testing a stale agent-discovery state.
- Invoke explicitly. For a repeatable review, use:
copilot --agent security-auditor --prompt 'Review src/auth and tests for high-confidence authorization vulnerabilities. Do not modify files.'- Inspect the result. In an interactive session, use
/diffto inspect changes and/reviewto analyze staged or unstaged changes. A review agent that was intended to be read-only should have no unexpected working-tree modifications.
A reliable remote-delegation workflow
- Run
git statusand inspect both the diff and untracked files. - Confirm that the repository’s GitHub remote, branch, and permissions are correct.
- Decide whether current local changes should be committed as the delegation checkpoint. Do not accept that step without reviewing what it includes.
- Confirm that cloud agent is enabled for the account, organization, enterprise, and repository, and that Actions and AI-credit capacity are available.
- Write a prompt with acceptance criteria, test commands, scope limits, and expected documentation.
- Run
/delegateor&. - Follow the session and review the draft pull request as you would a junior developer’s contribution.
- Check the complete diff, tests, dependency changes, generated files, permissions, and workflow modifications before approving or merging.
Local changes are not always required. An early delegation bug involving a repository with no local changes was fixed in the CLI changelog on November 3, 2025. The checkpoint question is part of the normal workflow when relevant, not proof that a nonempty local working tree is mandatory.
Choosing among CLI customization features
Custom agents are one part of Copilot CLI’s customization system:
| Need | Use |
|---|---|
| Rules that should apply to nearly every task | Custom instructions, such as repository coding, build, and test conventions |
| A named specialist with focused instructions or a restricted toolset | Custom agent |
| A repeatable procedure with scripts, resources, or reusable guidance | Skill |
| Deterministic validation, logging, scanning, or lifecycle automation | Hook |
| Access to a database, ticketing system, cloud platform, or internal API | MCP server |
| Local unattended execution | Autopilot |
| Background work delivered through GitHub review | /delegate |
A custom agent should not be used to encode a rule that every agent needs, and model instructions should not be used as a substitute for a deterministic hook when a policy must be enforced.
Practical profile patterns
Use the security auditor above as a template. Other useful profiles include:
- Test runner: allow repository inspection and command execution, name the permitted test commands, require it to report failures and avoid unrelated refactoring.
- Documentation maintainer: restrict scope to documentation and examples, require links and code samples to be checked, and prohibit production-code changes.
- Database migration reviewer: inspect migration ordering, rollback behavior, locking, data loss, and compatibility; make it read-only unless implementation is explicitly requested.
- API contract reviewer: compare routes, schemas, generated clients, tests, and documentation; require evidence for breaking-change claims.
For every profile, specify five things: the files or directories it may inspect, the tools it may use, the actions it must not take, the checks it should run, and the structure of its final report. This makes the agent easier to evaluate and less likely to wander into an unrelated task.
Troubleshooting
The custom agent does not appear
- Confirm that the file is inside
.github/agents/or~/.copilot/agents/, or another documented discovery location. - Confirm the filename ends in
.agent.mdor, for CLI versions that support it,.md. - Check YAML frontmatter syntax and ensure
descriptionis present. - Make sure the current directory is inside the intended Git repository.
- Restart Copilot after creating or changing the profile.
- Check for another profile with the same identifier. Because GitHub’s documentation conflicts on user-versus-project precedence, use a unique filename rather than trying to win a shadowing contest.
- Temporarily remove fields that may not be supported by the installed CLI, especially cross-surface fields,
infer, model settings, or MCP configuration.
The agent appears but uses the wrong tools
If tools is omitted, the default is broad access to available tools. Add an explicit list such as:
tools:
- view
- grep
- globAlso review whether an inherited or shadowing profile with the same name was loaded. Saying that an agent should not edit files is not equivalent to removing edit tools.
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 matchThe right agent is not selected automatically
Inference depends on the model and the profile description. Add concrete task types, trigger phrases, directories, file types, and exclusions. For a security or release gate, bypass inference and use /agent or --agent explicitly.
/delegate is unavailable
Check authentication, the repository’s GitHub connection, repository permissions, cloud-agent entitlement, organization and enterprise policies, AI-credit and Actions-minute limits, and managed-user restrictions. The CLI can be available while cloud agent is disabled. If the problem is policy-related, an administrator may need to enable the relevant cloud-agent setting.
Unexpected local changes are included
Inspect git status, staged and unstaged diffs, and untracked files before accepting the checkpoint request. Delegate from a clean or intentionally prepared branch when the boundary matters.
The cloud task cannot finish
Typical causes are missing setup commands, dependencies that require unavailable secrets or services, workflows awaiting approval, exhausted Actions minutes or AI credits, an underspecified prompt, or a custom agent that assumes local credentials and tools. Put setup instructions in the repository or the appropriate cloud-agent environment configuration, and never paste secrets into the request. A draft pull request that reports success still requires human verification.
Recommended Free Tools
Review checklist for generated work
- Does the diff stay within the requested scope?
- Did the agent change dependencies, CI workflows, permissions, or deployment settings?
- Do tests cover the new behavior, and were they actually run in the relevant environment?
- Are failures caused by the implementation, missing services, secrets, or an incomplete setup?
- Does the code handle authentication, authorization, input validation, errors, concurrency, and rollback correctly?
- Did generated documentation and examples receive the same review as production code?
- Were secrets, tokens, private URLs, or sensitive repository contents exposed in logs or the pull request?
- For a security review, is every finding supported by concrete evidence rather than a generic warning?
Frequently Asked Questions
Can I close my terminal after using Copilot CLI delegation?
Yes. A task sent with /delegate or & runs in Copilot cloud agent, so it can continue after the local terminal or computer is shut down. The result is normally a remote branch and draft pull request. Local autopilot is different: it depends on the local process continuing to run.
Does every Copilot plan support /delegate?
No. GitHub documents Copilot CLI as available with all Copilot plans, subject to policy, but cloud-agent delegation has separate paid-plan or organizational-entitlement and repository-policy requirements. An administrator can disable cloud agent even when the CLI itself works.
Can I force a custom agent when using /delegate?
The current delegation documentation does not document a /delegate --agent NAME option. Select custom agents deterministically with /agent or --agent for local CLI work. For cloud work, select the custom agent through GitHub’s Agents interface or issue-assignment workflow rather than assuming the locally selected agent is inherited.
Why should a read-only custom agent have a restricted tools list?
Instructions such as Do not modify files are model guidance. Removing edit, shell, or other unnecessary tools provides a stronger operational boundary. For example, a review agent can use view, grep, and glob without receiving the tools needed to change the working tree.
How can I control Copilot CLI usage?
Use /usage to inspect session usage and set a soft response limit with --max-ai-credits or /limits set max-ai-credits. The actual cost depends on model, tokens, cached context, and task duration; cloud-agent work also consumes GitHub Actions minutes.
The Bottom Line
Install or update Copilot CLI, create a reviewed .github/agents/*.agent.md profile when you need a repeatable specialist, and invoke it explicitly for important local work. Choose local autopilot only when you accept autonomous changes on your machine. Choose /delegate or & when you want Copilot cloud agent to continue remotely and deliver a draft pull request. Treat the two agent systems as separate: verify cloud-agent entitlement and policy, do not assume local custom-agent settings are inherited, and review every generated diff before merge.
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.
