Skip to main content

🌐 Semgrep AppSec Platform

Added

  • Project archiving is now in public beta for organizations on the default tenant. Archive a project to stop scanning it and make its findings read-only without deleting its data. Semgrep also treats a project as archived when its repository is archived in GitHub, GitLab, or Bitbucket Data Center. Open findings from archived projects are hidden from the default views of the Findings page, the Dashboard, and reports. For more information, see Archive a project.
  • The hosted Model Context Protocol (MCP) server for Semgrep AppSec Platform can now filter projects by tag and edit projects, including their tags. You can only edit projects that you have permission to manage. If your organization uses Teams to manage project access, the MCP server can also list, create, update, and delete teams. For more information, see Tools and capabilities.
  • Audit logs now show when an action was taken through the hosted MCP server. They also name the MCP tool and client that took the action. MCP tool calls that change data now write an audit event even when the underlying action didn’t write one before. Tool arguments aren’t recorded.
  • Rule-based Semgrep Code findings now have Open in editor, and a finding’s rule modal has Edit rule. Both open the rule in Semgrep Editor with the finding’s file as the test code, scrolled to the finding’s line. Semgrep doesn’t save the finding’s code as the rule’s tests unless you click Use as test code. For more information, see Write rules using Semgrep Editor.

Changed

  • The Library in Semgrep Editor now loads rules in pages instead of downloading every rule each time you open it. You can filter rules by language, source, severity, visibility, and whether you created them, and group them by folder or ruleset. Custom rules appear in their folders. Rule runs in Editor also show their first results sooner.
  • The hosted MCP server now offers fewer, consolidated tools, and its tool descriptions no longer refer to HTTP routes. Any MCP client can now run scan and workflow jobs with the semgrep_run_job tool, including clients that don’t support MCP tasks. Validation errors now name the argument or tool that caused them.
  • When you connect GitLab for Semgrep Managed Scans, the GitLab Self-Managed option is now GitLab Self-Managed or Dedicated. Select this option for GitLab Dedicated and enter your instance URL. The GitLab Cloud option connects to gitlab.com. For more information, see Add a GitLab repository to Semgrep Managed Scans.

Fixed

  • Fixed an issue where a policy that combined the Dependency is malicious condition with a Branch type condition set to Non-primary branch couldn’t be saved. Malicious dependencies are detected on non-primary branches, so these policies can now be saved.
  • Fixed an issue where Settings > Usage double-counted seats from expired license terms. For example, two consecutive 100-seat terms showed as a 200-seat limit.

💻 Semgrep Code

Added

  • Semgrep can now write its scan log as JSON as well as text. Set the SEMGREP_LOG_JSON_FILE environment variable to a path, and Semgrep writes one JSON object per line for each phase start and finish, warning, and error, each tagged with the scan’s details. Use it to find why a scan failed or slowed down with tools like jq, or to send scan logs to your log pipeline. The text log doesn’t change.

Changed

  • Semgrep Code is less likely to run out of memory on large projects with cross-file (interfile) analysis. Previously, memory could jump sharply at the end of indexing, and the scan could be killed before it reported any findings.
  • Semgrep Code now follows calls to PHP methods that come from traits, so it finds more issues in code that uses traits. Traits combined with as or insteadof aren’t supported yet.
  • Go parsing now supports more syntax:
    • Generic type aliases now parse, and existing alias patterns can match them. Rule patterns still can’t spell out generic parameter lists.
    • Signatures that mix unnamed map parameters or results with other types, such as ([]string, int, map[int]int, error), now parse correctly, so function patterns can match them.
    • Chained indexing such as table[key][next] is now represented as nested index expressions, so nested-index patterns match again.

Fixed

  • Fixed an issue where Semgrep logged credentials embedded in a proxy URL, such as HTTPS_PROXY=http://user:password@proxy:8080, when it reported proxy settings.
  • Fixed an issue where logs and traces included snippets of scanned source code when a file failed to parse. Semgrep now reports the file, line, column, and length of the unexpected text instead. Dependency resolution errors for malformed lockfiles are reported the same way.

⛓️ Semgrep Supply Chain

Added

  • Semgrep Supply Chain now reports findings for Python dependencies that a requirements.txt file declares with a version range, such as litellm>=1.10, or without a version, such as flask. A finding appears when the declared range overlaps an advisory’s affected versions. Because Semgrep can’t confirm the installed version, these findings are labeled Needs review. A range next to a pinned copy of the same package doesn’t produce its own finding. This requires Semgrep CLI version 1.184.0 or later, which Managed Scans already uses, and doesn’t require --allow-local-builds. For more information, see Supply Chain requirements and feature support.

Changed

  • In AppSec Platform, Supply Chain now says “threat” instead of “incident” in the threat widget on the Dashboard and the Advisories page, the advisory details page, the advisories table, the Advisories and Dependencies filters, notification policies, and Slack notifications. For more information, see Supply Chain threat notifications and response.
  • When a Supply Chain scan has errors, the Supply Chain results may be incomplete alert in its scan details on the Projects page now shows a message for each type of error, such as Lockfile parsing failed or Dependency resolution failed, instead of one generic warning.
  • With --allow-local-builds, when semgrep ci --no-suppress-errors exits with code 2 because dependency resolution failed, the reason now also appears in the errors field of the JSON output. Each entry names the failed subproject’s resolver or lockfile error. Resolver errors include the failed command and the end of its output, with common credential formats redacted.

Fixed

  • Fixed an issue where the Yarn v1 lockfile parser reported a package named dependencies with an empty version when a blank line came before an entry’s dependencies block. The parser now keeps that entry’s dependency relationships, and reports entries without a version line as parser errors.
  • Fixed an issue where Gradle dependencies were attributed to the root manifest when a module’s project name didn’t match its directory, for example a renamed include or a remapped projectDir in settings.gradle. Dependencies are now attributed to the module’s own build file.

🤖 Semgrep Multimodal

Added

  • You can now write memories that apply only when Semgrep Agentic Workflows triages server-side request forgery (SSRF) issues. When you add a memory in Rules & Policies > Memories, select SSRF under AI Detection in Rules. Available on paid plans and trials. For more information, see Triage with memories.

🛡️ Semgrep Guardian

Added

  • On the Policies page, the Semgrep Code rules list now includes rules that run in Semgrep Guardian but aren’t in your Semgrep Code policy, labeled Guardian only. These include Guardian’s default rules and rules enabled with Enable in Guardian scans in Editor. Turning on CI for one of these rules adds it to your Semgrep Code policy.

🔧 Semgrep Community Edition

The following versions of Semgrep Community Edition (CE) were released during the week of November 9-15, 2026:

1.184.0