feat(ai): model selector, chat sessions, markdown rendering, stop generating

[exitzerolabs-admin]

Summary

• Model selector: configurable LLM model per provider (Claude Opus 4 / Sonnet 4 / Haiku 3.5, GPT-4o / GPT-4o Mini) via AI settings
• Chat sessions: persistent per-file sessions with create/switch/delete, localStorage-backed, auto-titled from first message
• Markdown rendering: react-markdown + remark-gfm for AI responses with Tailwind-styled components
• Stop generating: AbortController (browser) + Arc (Rust) cancel mechanism with stop button + Escape key
• Enhanced system prompts: deeper STRIDE expertise, canvas positioning guidance, position data in model context, markdown formatting instructions

Code quality fixes

• cancel_chat_stream returns Result<(), String> per project convention
• Runtime value imports for session limit constants (no magic numbers)
• migrateSessionKey wired up to prevent session loss on Save As
• XSS safety comment on MarkdownContent (rehype-raw intentionally omitted)

Docs updated

• architecture.md, product-design.md, overview.md, market-analysis.md, releasing runbook, CLAUDE.md

Test plan

• npx biome check --write . — clean (1 pre-existing CSS warning)
• cargo clippy — clean
• npx vitest --run — 416 tests pass
• cargo test — 59 tests pass
• Manual: open a model, send AI chat message, verify markdown renders
• Manual: switch sessions, create new session, delete session
• Manual: change model in settings, verify it's used for next request
• Manual: click stop button mid-stream, verify partial response kept
• Manual: press Escape mid-stream, verify stop works
• Manual: Save As on unsaved file, verify sessions migrate

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
threat-forge	Ready	Preview, Comment	Mar 4, 2026 0:24am