No findings match the current filter.
Independent multi-agent review · main @ 52cc9e2
Three AI agents swept the Goldfinch codebase in parallel; every finding below was then independently re-traced against source before it was written down. The report is organized the way the codebase thinks about itself — by trust boundary. Nothing here is speculative pattern-matching: each card carries the exact file, the offending lines, a concrete trigger, and a suggested fix.
Bugs cluster at the seams between trust domains. Click any dot to jump to its finding.
src/shared decision modules, the settings store's validate / normalize path, and renderer focus / tab-type / dispatch state machines.The three agents ran read-only and in parallel against a live dev:automation instance. Their raw output was not trusted verbatim — an orchestrator (Claude Fable) re-opened every cited file and traced each data flow end-to-end, promoting a finding only when the code path was unambiguous. Several agent claims were downgraded or reframed in this pass (for example, the page-context editFlags gap is real but the guest re-checks edit state at execution, so it is filed Low rather than Medium).