New CodeRabbit Features: Security Agents for Multi-Repo Deep Scans

The window between a vulnerability becoming known and someone exploiting it is shrinking. Attackers now use AI agents to find weaknesses in software faster than teams can review it. The defense has to work the same way, with AI agents that find and fix those vulnerabilities before an attacker does.

We launched CodeRabbit Secure in August to go after application-specific vulnerabilities. Its agents are purpose-built for security and use semantic analysis to hunt for them the way an attacker would, then help developers fix what they find.

Today, we’re releasing new CodeRabbit Secure features. Deep Scan now follows a risk across linked repositories, and runtime verification replays an exploit attempt in a throwaway sandbox to confirm which vulnerabilities are real. Security Check stops a pull request from merging while a critical issue is open, and new integrations carry findings into the ticketing systems teams already use.

CodeRabbit reviews every pull request diff for security vulnerabilities when the PR is opened. A Deep Scan extends that security review to the code already in your repository, tracing attack paths and uncovering risks that predate the latest diff. As changes merge, automated Delta Scans revisit changed code and areas where earlier verification has become stale, keeping security coverage current without repeating a full scan.

Deep Scan examines committed code throughout a repository and establishes a security baseline by identifying risks that shipped months or years ago. Ongoing PR security reviews monitor each incoming change, grading what they find for reachability and exploitability so the priorities in a pull request reflect real risk. Blast Radius and Architecture Review in Change Stack extend that view beyond the changed lines, showing the callers, downstream consumers, and trust boundaries a change could affect.

The Attack Surface Map ties the two together. It records which security-relevant areas of a repository have current verification and which ones a merge has made stale, so you know what’s covered and when it’s time to scan again. Together, these capabilities keep the security picture connected to the code as it changes.

None of this replaces the static checks already in your stack. CodeRabbit Secure works alongside them, using semantic analysis to answer the questions fixed rules can’t, like whether an authenticated user actually owns the data they’re requesting, or whether another code path can bypass that check. Those are the questions attackers now put AI agents to work on, and CodeRabbit Secure puts agents to work on them for you.

The new release closes the gap between a single scan and a codebase that keeps changing. Deep Scan now sees more of the application before reporting a finding, and teams have more ways to act once it does.

Production systems often span multiple repositories. When a request crosses a repository boundary on its way to sensitive code, a scan limited to one repository may miss where the request originated or which safeguards apply. That missing context can make protected code look vulnerable or hide a real attack path.

Deep Scan now reads linked repositories to trace requests across services and libraries. Before publishing a finding, it accounts for safeguards elsewhere in the application and evaluates the potential issue in the context of the service where it appears. That cuts noise and keeps your team focused on real issues. This is especially helpful for teams running microservices or shared libraries, where the risk often sits across a repo boundary.

If CodeRabbit has already discovered related repositories during code review, Deep Scan uses the same links, and discovery continues as you merge. You can also link repositories by hand and add a note on how they relate. Either way, you choose which linked repositories a scan may read. The one you selected stays the target, and the others supply context without receiving a scan of their own.

Linked repository context is available for Deep Scan on GitHub Cloud, GitHub Enterprise Server, and GitLab Cloud. A scan can read up to 20 linked repositories, subject to plan limits. The docs cover setup.

Every Deep Scan finding already passes through a verification agent before it is published. The verifier reopens the cited code, checks reachability, looks for safeguards elsewhere in the application, and rejects anything the evidence does not support.

Runtime verification adds another form of evidence. For eligible findings, CodeRabbit replays the exploit attempt in a throwaway sandbox, never against your production systems. Instead of a theoretical risk, you get steps to reproduce the finding and evidence from the run showing whether it holds up, so you can see the path from vulnerability to exploit and decide how to respond. Security engineers can triage faster, and developers can see which findings are worth fixing first.

Runtime verification is available today for Deep Scan findings.

A full Deep Scan reads the whole repository, which is the right way to set a baseline. Delta scans start from that baseline and rescan only what has moved since, meaning files added or changed, plus the parts of the Attack Surface Map those changes made stale. Code nobody touched keeps its verified status, because re-checking it adds cost, not security.

You pay only for the work that matters, at a fraction of the cost of a full scan. Set a verified-coverage target, and CodeRabbit runs a delta scan whenever coverage dips below it, within your monthly spending cap, so coverage keeps pace with your code and nobody has to start a scan.

You can also recheck any finding against the current tip of the default branch. If a finding has already been fixed, a recheck confirms it. If its severity is off, you can adjust it and optionally save that adjustment as a learning for future scans. And when a fix merges, CodeRabbit now verifies it and resolves the finding automatically, so fixed findings don’t linger.

You also decide how much a scan can spend. By default, Deep Scan runs until it reaches full coverage. If you set a spending limit, the scan stops when it hits that budget. If that happens before it finishes, you can resume the scan and continue from where it stopped, with no need to start over. These controls help security and engineering leads keep findings accurate and costs predictable, so teams spend their time on the issues that matter.

CodeRabbit Secure now lets repository admins block merges when a pull request has open critical security findings. It runs as a native GitHub check and blocks the merge until CodeRabbit verifies the fix for critical vulnerabilities.

In the CodeRabbit app, go to Security → Repositories, select a repository, open Settings, and enable Publish CodeRabbit Security check. By default, the check fails if a visible Security review comment is rated Critical, though you can choose a different severity threshold. A failed check blocks merging only if a repository administrator adds CodeRabbit Security as a required check in a branch protection rule or ruleset.

The check report shows comment counts by severity, verified fixes, and the remaining blockers, with each linked to its comment. Marking a GitHub thread as resolved doesn't clear a blocker. It persists across pushes until CodeRabbit verifies the fix, so reviewers always know what’s still blocking the merge.

Security Check works on GitHub today, and other platforms are on our roadmap. More details are in our docs.

CodeRabbit now gives engineering and security leaders organization-wide reports on PR security reviews, with filters and exports for Deep Scan findings. It shows whether findings are actually getting fixed and where risk is piling up. Under PR security reviews, the report includes:

For codebase findings, teams can filter Deep Scan results by reachability and exploitability to narrow the list to what needs attention first, and export the filtered list as a PDF.

Once your team decides to address an issue, the work needs to reach the team responsible for the fix. CodeRabbit lets you preview a Jira or Linear ticket from any Deep Scan or PR security finding and create it with one click. You can also set a severity threshold and have CodeRabbit create tickets automatically, eliminating the need to copy findings into the backlog. This helps security teams hand findings straight to the people who own the fix.

Security teams often need to analyze findings alongside data from other systems, like asset inventories, incident history, and compliance reporting. For customers on CodeRabbit’s Enterprise plan, the read-only CodeRabbit REST API returns Deep Scan findings so you can export them into your data lake, SIEM, or your own internal agents. All other CodeRabbit Secure features are available on Advanced and Enterprise plans, including the 14-day Advanced trial.

curl --request GET
--url https://api.coderabbit.ai/v1/security/scans/code
--header 'x-coderabbitai-api-key: <api-key>'

Run a Deep Scan on your code today

With this release, CodeRabbit Secure carries verified findings from committed code and incoming changes through to the decisions that determine what ships. It’s what we mean when we say CodeRabbit is the governance and control layer for agentic software development, enabling your teams to trust the code they ship while moving at agentic speed. AI will find the exploit. CodeRabbit helps you find and fix it first.

In the last 30 days alone, CodeRabbit Secure found more than 8,000 high and critical vulnerabilities across 600 organizations. The only way to know what’s in your code is to look, and from now until November 8, CodeRabbit is making $1 million in Deep Scan usage available at no cost. Organizations with two or more members, including current, former, and new customers and open-source teams, can access up to $5,000 on a first-come, first-served basis while the allocation lasts.