Faster whole-project audits and fixes for strict MCP clients.
Fixed audits of file contents with no project root, which failed with an internal error once compiled scrolls were installed. They now run on the new engine, with the submitted files as the whole project.
Fixed audit results that MCP clients rejected for not matching the tool's output schema while SDK modules were still being prepared, or when a diff did not apply.
Removed the obsolete startup override for per-file audits. Rule concurrency now allows 1.5 times the available CPU cores.
Whole-project audits now merge file selections before evaluation and run rules concurrently according to available CPU cores.
Fixed a crash when whole-project audits prepared shared project and SDK data concurrently.
Audit continuation tokens now survive failed reads and keep their connection alive while results remain available.
Makes auditing changes much faster and keeps long audits from timing out.
Auditing changes is much faster: audits of edited files no longer slow to a stall in large projects.
Long audits now keep the agent informed: a call that is still running gets a reply within 50 seconds saying findings are on the way, instead of timing out.
Fixed a crash in Settings → Projects when the project list was read by VoiceOver or other accessibility tools.
Fixes audits in large projects stalling with rules still running.
Fixed audits in large projects stalling with rules still running: rules that reach many declarations now prepare them in batches the integration helper accepts.
Stored audit results in large projects are no longer dropped by the integration helper.
Paged whole-project audits, faster evaluation and the SwiftUI Layout & Resizability scroll.
Whole-project and folder audits return results in pages while the audit runs: agents fetch the next page with read_findings until none remain, and can ask again for a lost page.
Long audits keep their results until they finish, instead of losing the rest after ten minutes.
Audits are faster.
Guidance now appears when any of your audited targets can use it, so projects with several platforms see more findings.
A rule that appears in more than one installed scroll is reported once.
Suppression comments and disabled rules apply to all findings, and agents are no longer asked for context about suppressed findings.
New SwiftUI Layout & Resizability scroll.
Updated Proper SwiftUI, Modern Swift and Performant Swift Charts scrolls.
Scroll updates from 2026.10.2 onwards require this version.
Much faster audits, compact paged results, and framework types recognised without asking.
Audits are much faster: large multi-file audits finish in seconds rather than minutes, and independent batches in one audit run at the same time.
Audit results are compact and stay within agent tool limits, so they use far less of your agent's context. Each finding says whether it is definite or possible.
Agents can read several guidance entries in one request.
When an audit has more findings than fit in one result, agents page through the rest with read_findings instead of auditing again.
Fixed Apple SDK types not being recognised after updating from 2026.9.1, which left audits asking for framework types such as NavigationLink and CGFloat.
Framework types such as UIImage and SubscriptionStoreView are recognised through SwiftUI's re-exports and overlays, and audits that don't name a platform search every installed SDK, so agents are no longer asked for Apple types.
Fixed audits looping when a caller was in the same file as the code under review, and fixed results changing when a file was also sent as supporting context.
Project lookups search only the project your agent is working in.
Fixed agents occasionally failing to load SwiftFairy's tools when a session started.
Export Diagnostic Log now includes the helper's recent errors, including the reason an SDK module failed to load.
A saved license that can no longer be verified is now fetched again automatically. If SwiftFairy is offline, it tells you it couldn't verify your license.
The SwiftFairy plugins for Claude Code, Codex and Gemini update automatically.
This version reads a new Scroll format. Scroll updates from 2026.9.2 onwards require it.
Reliable SDK and project lookups, larger views, and a smaller SDK cache.
Audits now read Apple SDK types and your project's declarations correctly; previously these lookups failed and agents were asked for context SwiftFairy should have found itself.
Fixed audits failing in Claude Code when a file had more than five findings, or when project context was used.
SwiftFairy no longer interrupts you with prompts for missing code; your agent handles those requests.
Much larger SwiftUI views can now be audited.
The SDK cache is smaller and old versions are cleaned up automatically (about 12 GB on some Macs).
Bring Swift and SwiftUI code review expertise to your coding agents. SwiftFairy reviews code locally on your Mac and provides focused findings, practical guidance, and code examples from Nil Coalescing's knowledge Scrolls.