-
Document bounded, explicit replay of an emitted failed webhook using its original event ID and payload.
-
Clarify ambiguous acknowledgements, filtered/missing evidence, retention and independent delivery/history operations.
-
Add failure/retry/history regression coverage without changing production runtime, schemas, defaults or pricing.
-
Simplified the Console input flow to start with public URLs only.
-
Added optional, run-wide page-section and ignore-selector controls.
-
Grouped notifications, AI summaries, and operational limits into collapsible sections.
-
Refreshed the client-facing title, description, output labels, examples, and guidance.
-
Added transparent Store artwork and documented pay-per-event pricing.
-
Added safe custom page-check billing that stops after a client's run charge cap is reached.
-
Process checks in bounded batches so spend limits stop new work promptly while preserving input-order results.
-
Preserved all established input keys, defaults, accepted limits, history behavior, and the approved AI model override contract.
-
Made AI summaries request a strict structured response, validate every returned field, and fall back to heuristic analysis when the provider is unavailable.
-
Applied AI conclusions to small but meaningful changes while retaining the established ten-call safety limit.
-
Bounded AI prompts by UTF-8 bytes so multibyte page text cannot exceed the model-cost safety envelope.
-
Reserved the existing page-check charge before each AI request and avoided double-charging its emitted result.
-
Stopped AI work and preserved the prior snapshot when a run cannot reserve its next billable check.
-
Added response-size-aware concurrency without changing the established download-size input contract.