MulmoClaude v0.1.0 — first tagged release

GUI-chat with Claude Code — chat with Claude and get back not just text but interactive visual tools, persistent knowledge, and a growing library of skills.

✨ Highlights

• 9 specialised roles — General / Office / Guide & Planner / Artist / Game / Tutor / Storyteller / Musician / Role Manager. Each ships its own prompt + tool palette + sample queries; switching resets context.
• Personal wiki long-term memory — Claude builds and maintains a wiki/index.md + wiki/pages/*.md knowledge base that grows with every conversation. Cross-references via [[wiki link]] syntax.
• Skills (phase 0) — list and invoke any ~/.claude/skills/*/SKILL.md from the canvas via /<skill-name> slash-command. Project-scope override at ~/mulmoclaude/.claude/skills/.
• Charts — Apache ECharts plugin with bar / line / candlestick / sankey / network / heatmap. PNG export per chart.
• Documents / Spreadsheets / Forms / Mind maps / 3D / Music / HTML — rich plugin palette covering most "I want to make X" requests.
• Image generation — Gemini 3.1 Flash Image (nano banana) for generation + Ghibli-style transforms.
• MulmoScript storyboards — write multi-beat presentations with per-beat audio + image + movie generation pipeline.
• Docker sandbox by default — Claude Code runs in a sandboxed container with --cap-drop ALL, non-root, your ~/.claude mounted. Falls back transparently when Docker isn't available.
• Web settings UI — manage allowed tools and MCP servers from the browser (no JSON editing required). Auto-reload, no server restart.
• Web-based MCP server management — add Slack, GitHub, filesystem, etc. MCP servers via the settings UI.
• X (Twitter) tools — readXPost + searchX via the official X API.

🧠 Architecture & Infrastructure

• vue-router with history mode for deep-linkable session URLs (/chat/:sessionId, ?view=stack|files, etc.).
• Server-side session state — pub/sub channel keeps multiple browser tabs in sync without polling.
• Per-session pluggable MCP server — only the active role's plugins are exposed to Claude, keeping the tool list focused.
• Tool trace persistence — every Claude tool call/result is appended to chat/<id>.jsonl so reload restores the exact sidebar history.
• Wiki backlinks — pages auto-link back to the chat that created them.
• Auto-journal — daily summaries under summaries/ give Claude historical context without ballooning the context window.
• Structured server logger — log.{info,warn,error}(prefix, msg, data) with console + rotating file sinks.

🧪 Quality

• 1300+ unit tests (node:test) + 140+ E2E tests (Playwright Chromium).
• ESLint with cognitive-complexity gate (>15 = error).
• Cross-platform CI (Ubuntu / macOS / Windows × Node 22 / 24).
• TypeScript strict mode end-to-end.

🔒 Security

• Localhost-only bind (127.0.0.1) — no LAN exposure.
• CSRF guard on state-changing routes.
• Path-traversal-safe slug validation across sources, skills, wiki.
• Sandbox isolation for Claude CLI (Docker mode, default).

📋 Known limitations

• Skills in Docker + symlinked ~/.claude/skills — dangling links in the container; use DISABLE_SANDBOX=1 or flatten symlinks. README has the full workaround set.
• Conversation → skill capture (phase 1, #234) — landed on a feature branch but not in this tag. Will ship in 0.2.0.
• No Electron packaging yet — browser-only for now (#84 tracks this).

📚 Documentation

• README.md — installation, role catalog, skills, wiki, charts, X tools, settings UI, sandbox notes.
• docs/developer.md — env vars, scripts, sandbox layout, contributor guide.
• docs/logging.md — log levels, formats, rotation.
• docs/manual-testing.md — out-of-E2E smoke checklist.
• plans/ — design docs for every feature.

🙏 Thanks

Inspired by Andrej Karpathy's LLM Knowledge Bases idea and built on top of Claude Code.