Implement comprehensive artifact tracking to record all file modifications made by the AI agent during conversations. Add artifact persistence, garbage collection, and UI updates with new "Discuss" button styling.
Replace Claude Sonnet 4.6 with Sonnet 5 as default model, add Claude Fable 5, update OpenAI models to include versioned releases (GPT-5.5 Pro, dated variants), and update all references across enums, settings, tests, and documentation.
- Calculate height for consecutive assistant messages instead of just last message
- Track message role in ChatArea for proper height calculation
- Update MultiAgentIntegration test to use ChatMode enum
- Update officeparser browser bundle path and shim for v6
Introduce a new setting to control AI access to the web viewer tool and refactor prompt building to inject capability directives (memories, web search, web viewer) into all agent contexts. Update tool definitions to conditionally include web viewer and memory tools based on settings. Consolidate memories injection and active directives into a unified prompt builder pattern.
Refactor all test files to wrap tool response data in AIToolResponsePayload instances instead of passing raw objects directly to AIToolResponse constructor. Update assertions to access response data via payload.response property.
- Update error message to show actual max lines and words per line
- Return error message from updateMemories instead of success boolean
- Add SettingsService mocks to all agent tests to prevent initialization errors
- Update test expectations for new default search limits (30 results, 100 snippet size)
Replace the single Replan tool with three targeted alternatives:
ReviseStep, RevisePlan, and SkipStep. This gives the orchestration
agent more precise control over plan recovery — revising only the
current step, replacing all remaining steps, or skipping a step
entirely — rather than triggering a full replan each time.
Also give the orchestration agent read access to vault files so it
can resolve execution failures without routing back to the planning
agent, and update the execution agent prompt to stop it from
attempting self-recovery (gap resolution is now the orchestrator's
responsibility).
