🚀 What’s New
- AI Pentest Retest Management: You can now view a full history of retests for any AI Pentest run — including verdict summaries, phase progress, and remaining time budget — directly in the platform. A new cancel action lets you stop a pending or running retest at any time, and any reserved credits are immediately returned to your wallet instead of being held.
- AI Pentest Engine Settings Honored: When you configure specific engine options for an AI Pentest, those settings are now faithfully applied to the live scan — ensuring the test runs exactly the way you configured it every time.
- Live Credit Balance Updates: Your credit balance now updates in real time across every open tab. Whenever a credit movement occurs — a scan reservation, a remediation, a refund — the Dashboard reflects it instantly without polling, so you always know your available credits at a glance.
🛠 Improvements
- AI SAST Failure Diagnostics: When a scanning engine can’t complete a language analysis, the platform now records a clear, human-readable failure reason — such as a timeout or an agent issue — instead of a generic noresult message. You can see exactly what went wrong and why, directly in your scan report.
- AI SAST Smart Recovery: The scanning engines now gracefully handle cases where an AI agent finishes its analysis but returns results in an unexpected format. Instead of marking the target as failed, the engine salvages the findings it can — so you get results even from edge-case runs. Provider-side refusals are also detected automatically and retried on a fallback model.
- AI SAST Ignores Certificate & Key Files: Certificate and private-key files are no longer treated as code to scan, eliminating false partial-scan warnings on repositories that bundle TLS artifacts.
- Codex Remedium Failure Detail: When a remediation generation fails, the notification now includes the full diagnostic traceback — not just a bare error code — so your team can reproduce and resolve the issue without waiting on support.
- Codex Remedium Bulk Resilience: Bulk remediation runs that hit a total failure now close cleanly with a clear status and notification, instead of leaving the remediation stuck in a processing state.
- XBOM Import Error Messages: When an XBOM import exceeds the file-size limit, the import dialog now shows the specific size-limit message instead of a generic failure, so you know exactly what to fix.
- Sandbox Remediation Flow: Sandbox findings now pass through the remediation step without requiring an SCM connection, so the end-to-end partner demo runs uninterrupted from scan to fix.
- Organization Settings Validation: The Locations tab now refuses to save when the required Headquarters Location is empty, preventing silent data gaps in your organization profile.
- Connector Scope Guidance: Saving a Gitea, Forgejo, or Azure DevOps connector with a token that lacks the necessary read scopes now names the missing scope in a friendly message — instead of showing a raw server error.
- App Verification Deduplication: Domain verification is now atomic — two concurrent verification attempts for the same URL can no longer create duplicate entries, and the validated entry always takes precedence.
🔒 Security
This release includes a dedicated security hardening pass across connectors, access controls, and platform integrations:
- Azure DevOps SSRF Protection: Connector URLs for Azure DevOps are now validated against a strict public-host check — internal, loopback, and cloud-metadata addresses are refused before any outbound request is made
- Team Settings Access Control: The team settings page is now restricted to admin-role users only, closing an authorization gap that allowed any authenticated user to view or modify team configuration
- Pentest Webhook Integrity: Signed messages from completed, cancelled, or failed AI Pentest runs are now refused — a leaked webhook secret can no longer alter the status of a finished run
- TOTP Credential Handling: AI Pentest retests that require TOTP authentication are now safely refused before any credits or compute are consumed, until an engine can fully support the login flow
- Plan Price Integrity: Plan catalog prices are now validated to reject negative or non-finite values, ensuring billing integrity across all subscription tiers
🔧 Fixes
- Sandbox repository writes: Saving a repository in the sandbox no longer fails with an internal error — the write completes successfully and the sandbox scope updates as expected.
- Self-hosted connector tokens: A mistyped Gitea or Forgejo token now shows the friendly checkyourtoken message again instead of a raw HTTP error line.
- Codex Remedium stuck remediations: A remediation that fails during issue generation no longer stays permanently locked — the document is now marked with a terminal failure status and can be deleted without a conflict error.
- Deep Analysis snippet extraction: Findings that reference archived or binary files no longer trigger processing failures — the engine skips the snippet gracefully and continues scanning.
- Workflow PR creation: Transient network failures during automated PR creation now retry automatically instead of leaving a repository permanently without its required secrets.