Skip to main content
AI Development

Copilot app security extension: what to allow first

GitHub's Copilot app security extension explains and fixes code scanning alerts in your local copy. Fixes aren't pushed, and dismissals still reach GitHub.

Sean McLellan profile photo

Sean McLellan

Lead Architect & Founder

5 min read
Editorial illustration of a small wooden repair bench where a brass magnifying loupe on an articulated arm hangs over an open brass clockwork movement, with a tray of loose gears and copper tools beside it.
Constructed diagramCareful repair on one local bench before anything leaves it. Illustration, not GitHub product UI.

A security extension for the Copilot app entered public preview on Oct. 28, 2026. It lists a repository's open code scanning findings inside the app, explains or patches the ones you select in your local working copy, and adds a /security-review command for changes you haven't pushed yet. For Business and Enterprise administrators, the decision is whether to allow the github-security plugin and for which teams.

The main limit is what a patch does. The docs say Fix "changes your local working copy only" and does not commit, push or open a pull request, and a local edit does not close the finding. It closes only after the change reaches the affected branch and the scan runs again. Dismissing is different: it updates the server right away.

What does the GitHub security extension for the Copilot app do?

It puts scan findings and a pre-PR security review into the same session where developers already work. According to the changelog, the extension adds a Security canvas where you can:

  • browse every open finding, ordered by severity, with search and filters
  • open a finding's details and affected lines, or jump to its full page on the website
  • select one or more and click Explain or Fix
  • see pull requests already linked to an item, so two people don't patch the same thing
  • dismiss items you don't plan to address
  • run /security-review on your current changes and see the results next to existing findings

The documentation page adds one scope detail: the extension shows static analysis results only, "not Dependabot or secret scanning alerts."

Two-column diagram titled GitHub security extension: what stays local. Left card, Stays in your working copy: Explain, Copilot explains selected alerts; Fix, changes local files only, no commit or push; /security-review, checks your current changes. Right card, Updates GitHub: Dismiss updates the alert on GitHub; Delegated dismissal sends a request for approval; the alert resolves after the fix reaches the branch and code scanning runs again. A band notes that Explain, Fix and /security-review use AI credits and browsing and dismissing do not.
Constructed diagramConstructed diagram from the GitHub changelog and GitHub Docs, Oct. 28, 2026. Not GitHub product UI.

Fixes stay local; dismissals reach GitHub

The extension splits its actions into two groups, and that split decides how much review it needs.

Local actions. Explain, Fix and /security-review run in the agent session on your machine. Fix edits files and stops there. You review the diff, run your tests, then commit and push through your normal review process. The docs call the local review "a focused review of your changes, not a full repository security audit or a replacement for code scanning," and its results are stored separately from the scan's list.

Actions that change the server. Dismissing closes the item on the website, with a reason and a comment. If your organization uses delegated dismissal and your role needs approval, the extension submits a dismissal request instead, and the item stays open until an authorized reviewer approves it.

The extension's tools also follow the app's tool permission settings, so if your developers must approve tool use today, the agent asks before using these tools too.

What do you need to use it?

Less than you might expect for the local review, and more for remediation. The requirements:

Scroll sideways to see all 2 columns.

What you wantWhat it needs
Install the extension and run /security-reviewAny Copilot plan and the app. No scanning setup or Code Security license needed
Business or Enterprise seatsAn administrator must enable the app policy; enterprise managed settings can control which plugins developers install
See, patch and dismiss existing findingsCode scanning enabled, permission to view the results, and Code Security for private repositories
Hostinggithub.com or Enterprise Cloud with data residency. Enterprise Server isn't supported

Install it with copilot plugin install github-security@awesome-copilot, or in the app under Customize by searching for github-security. The extension is enabled automatically after install.

How does it compare with agentic autofix?

They attack the same backlog from opposite ends. Agentic autofix, generally available since Oct. 15, runs in the cloud: you assign a default-branch item (or up to 25 at once) to the agent, it reruns CodeQL to check its change and opens a draft PR. An earlier post covered how to scope that first assignment.

Scroll sideways to see all 3 columns.

Comparison dimensionSecurity extension (Copilot app)Agentic autofix
Where the change happensDeveloper's local working copyCloud agent
OutputUncommitted local changesDraft PR
How the item closesAfter the change reaches the branch and the scan runs againThe agent reruns CodeQL before opening the PR, which you then merge
StatusPublic previewGenerally available
Who enables itCopilot app policy and plugin settingsEnterprise "Agentic autofix" policy, then org or repo settings
BillingAI credits like any other promptCode Security AI credits and Actions minutes

BaristaLabs' interpretation: the extension fits findings a developer is already close to, such as issues in files they are changing this week. The cloud option fits a default-branch backlog nobody owns. Running both is reasonable, but the linked PR view matters, because an item assigned to the cloud agent can also be picked up locally.

Who should allow it first?

Teams that already scan, or that are trying AI Scan on pull requests, get the most from it, because the findings list is already there. Teams without scanning can still use /security-review, but the docs are clear that it isn't a replacement, so treat it as an extra check before opening a PR.

Hold off if your organization hasn't decided on the Copilot app itself. The app has its own policy and account settings; see how it can separate the license account from the repository account.

Why it matters for small engineering teams

For a small team, the cost of each finding is mostly the switch between the web page and the editor, and the time spent understanding an unfamiliar rule. The extension removes that switch and makes an explanation one click away. BaristaLabs' interpretation: that should shorten remediation time for routine issues, but it doesn't remove the review step, because the patch arrives as plain local edits and the item stays open until the branch is scanned again.

The cost is AI credits. Explain, Fix and /security-review "consume AI credits in the same way as any other prompt," using the model selected in the app, while browsing and dismissing don't use AI. No per-item cost is published.

BaristaLabs recommends allowing the github-security plugin for one team on one repository that already has scanning enabled. Keep tool approval on, turn on delegated dismissal if you want a second person to approve dismissals, and ask developers to rerun /security-review before each PR in that repository. After two weeks, compare items closed and AI credits used against the ones handled by the cloud agent or by hand, then decide whether to widen the plugin allowlist.

For help choosing between these paths, see AI consulting.

Sources and scope

These sources describe a public preview that is subject to change. They don't give per-item AI credit costs or quality figures, and BaristaLabs has not run the extension for this post.

Code scanning and Copilot

Plan a security extension pilot

Bring one repository with code scanning and your current alert backlog. BaristaLabs can help you decide which alerts go to the Copilot app and which to agentic autofix, set plugin and dismissal policies, and track AI credit use for the pilot.

One repository and one team is enough to start.

Turn this idea into a pilot

Which workflow should go first?

Use the readiness check to compare impact, effort, risk, owner, and next step before requesting a review.

  • 3-5 minutes
  • Deterministic score
  • No sensitive data
Check workflow readiness

Practical AI Workflow Notes

Want more practical AI operations ideas?

Get short notes on applying AI inside real small-business workflows — from document handling and customer follow-up to internal reporting, compliance, and automation guardrails.

A useful next step if you’re still exploring and not ready to request a 20-minute workflow assessment.

Occasional emails. Practical workflow guidance only. Unsubscribe anytime.