mulmoclaude 2.0.0 → 2.2.0

package/README.md

@@ -147,7 +147,7 @@ Recommended: ≥ 32 characters of random data (shorter values trigger a startup
 ## Roles, skills, and collections
 
 - **Roles** (sidebar selector): General, Simple (a much smaller prompt for light conversation), Office, Guide & Planner, Artist, Tutor, Storyteller, Settings. Each one biases Claude toward a workflow and surfaces its sample prompts.
-- **Skills** (`~/.claude/skills/<name>/SKILL.md`): personal skills shared across every project, plus project skills under `<workspace>/.claude/skills/`. Bundled "preset" skills (`mc-*`) re-seed on each boot.
+- **Skills** (`~/.claude/skills/<name>/SKILL.md`): personal skills shared across every project, plus project skills under `<workspace>/.claude/skills/`. Bundled "preset" skills (`mc-*`) re-seed on each boot. If installed Claude Code plugins carry so many skills that every request gets heavy, **Settings → Model** shows the count from the last turn and can stop loading `~/.claude` skills and plugins (the workspace's own keep working).
 - **Collections**: schema-driven data apps. Author your own (`data/skills/<slug>/schema.json` declares the model + UI), or use the Discover tab on `/collections` to import community collections from the official registry — or your own org / community registry by dropping `config/collections-registries.json` in the workspace. The Map tab draws the ontology graph across your collections, so you can see how their records reference each other, and the view header has a pulldown to jump straight to a related collection. A custom HTML view can put buttons on your records; give its `views[]` entry `"allowSendChat": true` and a press starts the work instead of leaving the prompt in the composer for you to send.
 
 ## Optional features
@@ -157,6 +157,7 @@ Recommended: ≥ 32 characters of random data (shorter values trigger a startup
 - **Download any workspace file** — the Files view previews text and renders known media; anything else (and any file too large to preview) offers a direct download of the bytes. This replaces relying on "Open in OS", which spawns a handler on the *server's* desktop — under Docker, WSL2 or a remote host there is none, so those files used to be unreachable from the UI.
 - **Local voice input** (macOS only, opt-in) — `whisper.cpp` for dictating chat messages without sending audio to a cloud API.
 - **Marp slides** — `marp: true` frontmatter on any markdown file renders a slide deck in the canvas with PDF export. Custom themes via `config/marp-themes/<name>.css`.
+- **Remotion scenes in MulmoScript stories** — with the optional remotion packages installed, ask for a Remotion animation on a story beat: the agent can write the scene component itself, so you can refine it in the conversation, or leave it to mulmocast to generate from a description.
 - **Auto memory** — the agent maintains a typed memory layout (`conversations/memory/<type>/<topic>.md`) and reads it ambient-style.
 - **Web Push on task finish** — enable in Settings → Notifications to get a push on your phone when the answer to a question you asked is ready, even with the browser closed. Requires the RemoteHost connection + a registered device (see [`docs/remote-host.md`](https://github.com/receptron/mulmoclaude/blob/main/docs/remote-host.md#web-push-on-task-finish-2086)).
 - **Google (Calendar / Tasks / Drive)** — link your Google account in Settings → Plugins → Google and consent in the browser; no Google Cloud setup needed. The agent then gets a `google` tool covering the full round trip — list, create, edit and delete calendar events and tasks (including putting a completed task back on the list), plus the Drive files it creates — and the phone remote can trigger the same commands. Events can be timed or **all-day**: ask for a whole day and you get one, rather than a midnight appointment.