When scans run
- First scan. When you connect a repository from the Repositories page, its first scan starts by itself: a full scan, which builds the repository wiki and the security baseline without any action from you.
- Rescans. When you connect repositories, you choose how often they are scanned again. You can change it for each repository at any time in the Scan cadence section of the Security page.
- Your own scans. You can start a scan of any connected repository from the Security page at any time.
A repository that joins your workspace from GitHub without going through the
Repositories page, such as a new repository in an installation that can
access all repositories, gets no first scan by itself. With After every merge,
its first merge starts its first scan.
Rescans after a merge
A rescan after a merge, or on a weekly, every-two-weeks, or monthly setting, looks only at what changed since the previous scan, and carries the rest of the previous result forward:- The code analyzers check only the files that changed. Findings in other files are carried over from the previous scan.
- Dependency checks list the dependencies again only when a manifest or lockfile changed. Otherwise the dependencies the previous scan found are matched again against current vulnerability data, so a newly published vulnerability in an unchanged dependency still shows up.
- The wiki regenerates only the pages of modules whose files changed. Links between modules are updated from the new code map without regenerating pages.
Scan types
A scan you start from the Security page runs on the repository’s default branch; a schedule can name a branch, tag, or commit.
Each scan runs on the exact commit it names, in the same isolated sandbox as
reviews. Scans never post to GitHub; results stay in the app. Each result keeps
its findings with their evidence and compares them with the previous scan of the
same repository: new, unchanged, fixed, or back again. An analyzer that could
not run is shown as not run, never as a clean result.
The wiki in pull-request reviews
On Pro and Team Pro, a pull-request review reads the wiki pages of the modules the pull request touches, with each module’s tests and its neighbours in the code map. The pages come from the repository’s latest completed scan of its default branch; nothing is generated for the review, and it uses no scan. The review treats the pages as background that can be out of date: it checks them against the code, and never reports a finding on the wiki alone. A repository with no completed scan is reviewed without its wiki.Schedules
On Pro and Team Pro a schedule runs every week, every two weeks, or every month, at the hour and in the timezone you choose, against a branch, tag, or full commit SHA. You can pause, resume, or delete a schedule; a paused schedule keeps its settings and history. Each scheduled run counts like any other scan. The schedule behind a repository’s weekly, every-two-weeks, or monthly rescan setting is managed from that setting on the Security page.Limits
- 20 new repositories a month per workspace, counted per UTC calendar month. A repository counts in the month of its first scan that did not fail. Rescans of a repository already scanned do not count. The first scan of a 21st new repository waits for next month and counts toward that month.
- Rescans are included, within a monthly cost cap for scans. First scans and rescans both count toward the cap, and rescans always leave part of it for the first scans of the new repositories the month still includes. When the month’s scans reach it, further scans wait until next month.
- Scan starts a month per workspace: 100 for first scans and full scans, failed scans included, and 1,000 for rescans that look only at what changed. Rescans always leave enough starts for the first scans of the new repositories the month still includes.
- 2 scans running at a time per workspace. Further scans wait in the queue, and a rescan after a merge waits while another scan of the same repository runs.
- Cost limit per scan. A scan stops if it reaches its cost limit. A security scan, or a smaller repository, finishes within it.
- Failed scans are not charged. A failed first scan does not use one of the month’s new repositories. The work a failed or stopped scan did still counts toward the month’s cost cap.

