Send

Three modes: **Plan mode (start here):** When the user asks for a document, call this first with intent only (a one-line description of what they're making). Returns tailored guidance plus the best-matching skills; no document is created. Then call CreateDocument again with html. **Copy mode:** Pass sourceShareId (share ID or URL of an existing document). The new document is a copy of that source. Do not pass html. **Create mode:** Pass html (see the html field for format requirements). Do not pass sourceShareId.

EditDocument modifies a document the user previously created or shared with Send. Use EditDocument whenever the user asks to change, tweak, fix, reword, restyle, add to, or remove anything from an existing document — even when they don't name Send or say the word "edit". In-flow requests that should route here include things like "make the heading bold", "change the title to …", "fix that typo", "make this paragraph shorter", "add a section about pricing", "change the button color", "swap the image", "update the date", or "delete the last slide". If the user is referring to the document you just created or were just discussing, this is the tool. Mechanics: edits via deterministic string replacement. REQUIRES at least one entry in `edits` — `instruction` alone does nothing. If you don't already have the document's current bytes, call GetDocument first. Then pass one or more edits — each replaces a unique substring of the current HTML. Edits apply sequentially against an in-memory copy; if any edit fails, none are persisted.

Unified image management tool. Controls what the user sees (upload area, image gallery) and what data is fetched (all images, specific images, or none). Common call patterns: - Fetch image data silently (no UI): `manage_images({ showGallery: false })` - Show specific images: `manage_images({ ids: ['abc', 'def'] })` - Show specific images + upload: `manage_images({ ids: ['abc'], showUpload: true })` - Upload only (no gallery): `manage_images({ ids: [], showUpload: true })` - Browse all: `manage_images()` — fetches and displays all recent images

Returns workspace images with proxy URLs for the gallery UI; each item includes an optional description for model context (not shown in the UI).

Creates a presigned upload target for the drop zone UI to upload directly to storage.

Confirms a direct-to-storage upload completed successfully.

Fetch an existing Send document by share URL or share ID. Returns the full HTML source and metadata; the response includes shareId for EditDocument or CreateDocument (copy mode).

Fetches skills by ID, lists available skills, or semantically searches them. Skills were previously called "guidelines", and users may also say "template" — when the user mentions making, editing, searching, or using a Send skill, guideline, or template, they mean these. Skills shape how documents are created — brand systems, layout patterns, doc-type playbooks. Workspace skills (created by this team) outrank Send skills (official, `send/*` ids) when both fit. **With id:** Returns the skill including its full content — pass an array of ids to load several in one call. Works for any skill you can access, including unlisted skills that never appear in lists or search. **With query:** Semantic search — describe what the user is making (e.g. "investor dataroom", "weekly metrics report") and get the best-matching skills grouped by source with similarity scores. For document requests, CreateDocument plan mode (intent) already runs this search; use query when the user explicitly asks to find or browse skills. **Neither:** Returns a list of all available skills (name and description only), workspace skills first.

Create, update, or delete a user-defined skill for this workspace. Skills were previously called "guidelines", and users may also say "template" — a user asking to create, edit, or delete a guideline or template means this tool. Call this only when the user explicitly asks for it — not during document creation. **create:** Provide name, description, and content. Before creating, load the skill-writing guide with get_skills(id: "send/skill-builder") — it covers naming, writing the description as selection criteria, and structuring content so the skill actually gets picked and followed. **update:** Provide id and any fields to change. **delete:** Provide id.

Embeds Send-managed content inline in the chat. For example: `"doc"` renders a published HTML document the user can read in place. Pass `"doc"` with the shareId after `CreateDocument` / `EditDocument` when the user should see the result. Requires the user to be signed in to Send.

Internal: re-fetches processed HTML for the doc currently rendered by show_component. Used by the widget's Refresh button. Not for direct model use.

Internal: persists in-place text edits made in the Send doc widget's edit mode. Position-addressed (text node index → byte splice); not for direct model use.