* feat: persist AI conversations and add chat history
Store transcripts and attachment previews in IndexedDB, separate conversation history from transport lifetime, and add document-aware switching with shared Storybook coverage. Preserve interrupted activity and guard stale writes and async switches.
* test: group chat history tests by domain
* feat: simplify chat history navigation and diagnostics
Use a compact header with searchable document-scoped history and a correctly anchored conversation menu. Move diagnostic copying into the menu with copy-result feedback, and separate Storybook fixtures from composition.
* fix: address chat history review findings
* test: cover harness shutdown and restored chat scrolling
* fix: preserve Portless proxy port in MCP routes
* docs: clarify UI animation conventions
* feat: configure reasoning display and animate disclosure
* fix: separate transcript following from reasoning disclosure
* fix: restore selected chat after document recovery
* refactor: separate chat history persistence and sessions
* style: format reasoning story imports
- Drive the app theme store from Storybook's theme toolbar\n- Use OpenPencil surface tokens for readable inline code in light mode\n- Verify light Shiki and inline-code contrast in the visual workshop
- Override inline code through the renderer component API instead of CSS cascade fights\n- Add light and dark Storybook states for prose, inline code, fenced code, streaming, messages, and attachments\n- Keep fenced Shiki rendering separate from compact inline tokens
- Keep the healthy Portless child when browser configuration is unchanged\n- Allow genuine configuration restarts enough time to register their replacement service\n- Cover configuration equality and delayed health readiness
- Classify authentication, access, model, rate-limit, network, and credit errors\n- Preserve the original provider error across the no-output stream consequence\n- Add optional toast actions for opening model settings\n- Consume handled chat send rejections to avoid global unhandled errors\n- Localize provider failure guidance and cover the authentication flow
- Move send orchestration, cancellation, Vision analysis, and optimistic history into a submission composable\n- Move composer image and node draft ownership into an attachment composable\n- Replace positional submit arguments with a named ChatSubmission payload
- Generalize transcript presentations across image and node attachments\n- Snapshot referenced nodes when sending so history survives graph changes\n- Reuse one attachment card and viewer for both attachment kinds\n- Localize attachment actions and remove image-processing details from the viewer
- Give configured profiles a full-width multiline selection surface\n- Show resolved model and provider metadata with a stable selection indicator\n- Keep Vision capability compact and link back to model-role settings\n- Split chat Markdown and profile themes into focused domain files
- Hold Vite startup and MCP restarts until the sibling service health endpoint responds\n- Tolerate transient Portless 404 responses during service registration\n- Cover worktree-origin CORS preflight and restart health polling
- Restore attachment inset spacing through the InputGroup theme slot\n- Mark curated multimodal models as Vision-capable and repair saved capability metadata\n- Default Vision to the Design profile when that profile accepts images\n- Use configured Design profiles in the composer instead of raw provider models\n- Surface missing Vision configuration directly in Settings\n\nCo-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Render node references and images with one attachment-card layout\n- Generate live previews for referenced scene nodes\n- Make the current-selection control toggle its active selection\n\nCo-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Import upstream Markdown CSS through the components cascade layer\n- Keep compact semantic overrides in a dedicated chat stylesheet\n- Preserve Shiki backgrounds through its supported theme variable\n\nCo-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Restrict assistant-rendered images to the active deployment origin\n- Cover local and self-hosted origins without hardcoded production hosts\n- Verify cross-origin, insecure, and data image rejection in browser tests\n\nCo-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Replace the fork-era Markdown integration with Comark and an explicit Shiki extension\n- Centralize OpenPencil theming, parser lifecycle, controls, and URL hardening\n- Remove Mermaid stubs and bundled math or diagram dependencies\n- Address contextual composer review findings and extend browser coverage\n\nCo-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Pin selected layers as bounded context without exposing metadata in transcript bubbles
- Show collapsible reasoning and copy individual assistant responses
- Autosize multiline prompts and cover the new chat states in browser tests
Co-authored-by: Jason Kneen <jason.kneen@bouncingfish.com>
- Reject open vector networks before contacting the provider\n- Cover closed and open vector targets explicitly\n- Label the stock photo prompt example code fence
- Reject text, line, structural, and populated container targets before provider requests\n- Preserve closed shape and Boolean geometry support without mutating target bounds\n- Cover accepted and rejected node families with focused tests
