Questions, answered
the way we work.

Published IDE extension releases. A report examines an exact release: requested permissions, capabilities that changed since the previous version, dependency additions, and the coverage of the evidence behind each finding. Findings are normalized into six scoring dimensions and mapped to a decision you can act on.

No. Website analysis inspects the published package without running it inside your editor. The local CLI can examine extensions already installed on your machine, and that analysis happens on your machine — nothing is uploaded unless you explicitly export a portable report.

They are the four report decisions. ALLOW means findings stayed within normal behavior. REVIEW flags changes that deserve human judgment. BLOCK marks behavior that should stop an install or update in governed environments. INCOMPLETE means we could not verify enough to decide — an honest outcome, not a pass.

In your browser. Report bundles dropped into Analyze or Reports are parsed and stored locally; the importer does not upload them. You remove them from the Reports library whenever you want.

You watch specific extensions. When a new release publishes, GuardRails compares its capability surface against previous versions and notifies you only about meaningful changes — quiet re-publishes without behavioral change do not page anyone. Notifications arrive by email, Slack, Jira, or a weekly digest depending on your workspace settings.

Analysis covers extensions published to marketplace registries used by editors such as VS Code, Cursor, and Windsurf. Guardrails is an independent reviewer — it is not affiliated with or endorsed by those platforms, and it does not modify your editors.

Scanning costs nothing and stays free: public exact-release reports, the registry, and the local CLI. Beyond that there are no prices yet — guided reviews of extensions you care about are free during launch with limited weekly slots, release monitoring is open to founding members through early access, and team pricing gets set together with design partners. Nothing pretends to be a checkout before billing exists.

Openly. A published benchmark runs the scanner against a reproducible corpus and states coverage and limitations alongside results, so you can see where detection is strong and where it is honest about gaps.

Both go through Contact. Report disputes should include the exact-release link and what you believe was missed; corrections ship as visible methodology updates. Suspected malicious extensions can be flagged for urgent review at security@abscissa.dev.