Security Agent

Security Agent scans a committed GitHub codebase for code vulnerabilities beyond the current pull request diff. Results appear in the CodeRabbit app under Security > Agent Findings, where you can inspect evidence, filter findings, export results, and start supported fixes. Use Security Agent when you want a repository-level view of application security risk. It runs against the repository’s configured scan branch, or the provider default branch when no custom branch is selected.

​

How Security Agent works

Security Agent runs a multi-stage pipeline rather than matching fixed rules against a single file or diff:

• Map — CodeRabbit builds a model of the repository so it can reason across files and services, not just within one file.
• Investigate — Agents examine likely vulnerable code paths and trace how data flows through the codebase.
• Verify — A separate verification step reopens the cited code paths and rejects speculative, test-only, dead-code, and unreachable cases before a finding is surfaced.

Deterministic scanners are used only as hints into this pipeline. Every finding is adjudicated against code evidence before it reaches you, which is why Security Agent can surface issues that rule-based tools cannot express — such as authorization bypass, IDOR, and business-logic flaws — with less noise.

​

What Security Agent scans for

Security Agent looks for practical code-level vulnerabilities across the selected repository. Findings can include issues such as injection, authorization bypass, IDOR, broken authentication, CSRF, sensitive data exposure, information disclosure, XSS, SSRF, open redirects, insecure deserialization, CORS issues, security misconfiguration, path traversal, weak cryptography, denial of service, LLM security risks, and other code security problems. LLM and AI application risks include prompt injection, improper output handling, and excessive agency. Each accepted finding is grounded in code evidence. When available, CodeRabbit shows:

• Category and severity
• Repository, file path, and line range
• CWE
• Description and recommendation
• Reachability: External, Internal, Unreachable, or Unknown
• Reachability call stack when CodeRabbit can trace a path from an entry point to the risky code
• Inline code context or a permalink to the scanned commit
• Status and fix status

Reachability affects how severity is presented. External and Unknown reachability leave the original severity unchanged, Internal reachability lowers severity by one level, and Unreachable findings are capped at low severity. Unknown means CodeRabbit could not determine the path confidently enough, not that the issue is exploitable.

​

Run a scan

You can start a Security Agent scan from the Security area in CodeRabbit. If one selected scan target is already running or blocked, other selected targets can still dispatch.

​

Scan status and activity

Each scan target moves through pending, running, completed, and failed states. A scan that runs for too long is marked failed with a message telling you it timed out or was interrupted, so you can run it again. The activity log records each run’s target, scan ID, status, duration, repository, who triggered it, and when. Completed runs link to that scan’s historical snapshot.

​

Configure repository scan settings

Repository security settings control which branch Security Agent scans, which code paths it skips, and whether scans run on a schedule.

​

Scan branch

For GitHub repositories, choose the branch CodeRabbit should scan. Provider default follows the repository default branch. Selecting the default branch clears the custom branch setting. If a configured branch is deleted, CodeRabbit warns that future scans will fail until you select another branch or return to the provider default.

​

Path exclusions

Use Path Exclude filters to skip code paths that should not count toward Security Agent scans.

• Add one repository-relative path or glob per line.
• You can add up to 100 excluded paths per repository.
• Each excluded path can be up to 512 characters.
• Absolute paths, Windows absolute paths, parent traversal, and control characters are rejected.
• Backslashes are normalized to /, leading ./ is removed, and duplicate entries are ignored.

The settings page also exposes path exclusions for Secrets scans. Dependency and SBOM path filters are not currently exposed in the customer-facing settings page.

​

Recurring schedules

Set a recurring full scan schedule per repository when you want CodeRabbit to scan on specific days and times. Schedules use 24-hour time, an IANA timezone, selectable days of the week, and selectable scan types: Code, Dependencies, SBOM, and Secrets. Each scheduled scan target dispatches independently. If one target is already active, CodeRabbit skips that target while other runnable targets can still start.

​

The Security area

The Security section in the CodeRabbit sidebar organizes the Security Suite into these pages:

• Overview — a dashboard of your latest security posture.
• Repositories — repository scan settings, including branch, path exclusions, and schedules.
• Agent Findings — code vulnerability findings from Security Agent.
• Dependencies — vulnerable dependency findings.
• SBOM — resolved component inventory and license visibility.
• Secrets Detection — leaked secrets and hardcoded credential findings.
• PR Findings — security issues raised during pull request reviews.
• Activity Log — a record of scan runs.

The Overview dashboard summarizes your latest known security posture across Agent Findings, Secrets, Dependencies, and SBOM. It shows:

• Summary cards with finding counts, affected repository count, and critical/high totals for each scan type.
• A Security Trend chart of findings by severity over time, filterable by repository and finding type across 30, 90, 180, and 365 day ranges.
• Latest Findings, the most recent critical and high Agent Findings with severity, category, repository, and file location.
• Highest-Risk Repositories, ranked by critical and high findings.

​

Work with Agent Findings

The Agent Findings page gives you a triage surface for code findings from completed scans. You can:

• Filter by repository, severity, category, and search text
• Open a finding drawer with evidence, recommendation, reachability, code context, and available actions
• Share a direct link to a finding
• Ignore a finding
• Export findings as CSV, JSON, or SARIF
• Start Fix with AI for supported vulnerability findings on GitHub repositories
• View a generated fix pull request when one exists
• Retry failed fixes

Ignoring a finding changes its status. If a later scan detects the issue again, CodeRabbit can reopen it. Fix with AI cannot start when another fix for the same finding is already pending or running. It opens pull requests for supported fixes; CodeRabbit does not merge those pull requests automatically.

​

Related Security scans

The Security area also includes other committed-codebase scan targets. You can start them from the same run dialog, but each has its own source and findings surface. Agent Findings can be exported as CSV, JSON, and SARIF. SBOM exports support CycloneDX JSON, CycloneDX XML, SPDX JSON, and CSV. Secrets Detection appears as a separate scan surface so secret triage stays separate from Agent Findings. On GitHub repositories, Fix with AI also supports eligible dependency and secrets findings in their respective scan surfaces.

​

Availability and limits

Security Agent is available for GitHub repositories. Repository security settings, scheduled scans, code context links, and Fix with AI pull request creation are also GitHub-only for this feature. New workspaces include a free trial with 10 code scans. Security Agent does not prove that a repository has no vulnerabilities. It reports findings that CodeRabbit can support with code evidence, and it preserves Unknown reachability when CodeRabbit cannot determine a path confidently enough. Code-scan credit estimates are estimates. They account for the effective scan branch and configured code path exclusions, but they should not be treated as a final invoice.

​

What’s next