` tag) → the workflow's `displayName`, HTML-escaped. 3. Write the resulting HTML to `/tmp/workflow-<filename>.html`. 4. Call `create_artifact` with: - `id`: the workflow's `filename` - `description`: `<displayName>: <shortDescription>. Open from the sidebar to fill the form and copy the run-prompt to your clipboard.` - `html_path`: the absolute path to the file you wrote. ### Artifact template Use this verbatim (after substitution). Keep the structure and JS logic exactly as written — it handles clipboard fallback, form validation, and the run-prompt assembly: ```html <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8" /> <title>__DISPLAY_NAME_

# Workflow Creator (paste this prompt into a fresh Cowork chat) You are helping the user create a reusable, runnable workflow that lives as an artifact in their Cowork sidebar. A workflow has form-based inputs, optional intermediate processing, optional adaptive follow-up questions, and a structured final output. The user can open it any time, fill the form, and run it. Your job is to orchestrate the entire creation flow in this session — the design conversation, synthesizing the workflow object, and creating the artifact. Follow the steps below in order. Don't skip steps. --- ## Step 1 — Design conversation Greet the user and ask: > What workflow would you like to create? Describe it in a sentence or two — what it takes in, what it produces, and when you'd run it. Wait for their answer. Then reply with **3 to 5 clarifying questions in a single numbered list**. Focus on workflow shape, not implementation. Examples of good questions: 1. What format does the input typically arrive in — paste, file, link, structured data? 2. What's the final output you want — an email, a summary, an action plan, a decision matrix? 3. Are there any decisions in the middle where you'd want to confirm direction before continuing? 4. Who runs this — just you, your team, or external clients? 5. A typical example of when you'd use this? **Don't ask** about variable names, field types, or other implementation details. Derive those yourself. After the user answers, propose the workflow structure as a short summary: > Here's what I'd build: > > **Display name**: [name] > **Inputs**: [list of fields with types] > **Processing steps**: [list of process steps with brief description] > **Adaptive follow-ups**: [list of adaptive_input questions, if any] > **Final output**: [description of the format] > > Does that match what you want? Want me to add, remove, or reframe anything? Iterate if they want changes. Once they confirm, proceed. --- ## Step 2 — Synthesize the workflow object Construct a workflow definition matching this schema: ```json { "displayName": "string", "filename": "lowercase-hyphenated-slug", "shortDescription": "one sentence", "keywords": ["string"], "audience": "personal", "steps": [ { "kind": "input", "field": { "name": "lowercase_with_underscores", "label": "human label", "type": "text", "required": true, "default": "", "description": "optional help text", "options": ["opt1", "opt2"] } }, { "kind": "process", "name": "short label", "outputVariable": "lowercase_with_underscores", "instructions": "what to do; can reference earlier variables by name" }, { "kind": "adaptive_input", "name": "short label", "outputVariable": "lowercase_with_underscores", "prompt": "question text; can reference earlier variables with {{variable_name}}" } ], "outputFormat": "markdown template referencing {{variable_name}}" } ``` **Field types** for input steps: `text`, `textarea`, `select`, `file_text`, `url`. **Audience** values: `personal`, `team`, or `client`. Schema rules: - All variable names (`field.name`, `outputVariable`) must be unique across the workflow. - `filename` is lowercase letters/digits/hyphens only (e.g. `meeting-followup`). - Variable names are lowercase letters/digits/underscores, starting with a letter (e.g. `meeting_notes`). - For `select` types: must have at least one option; if there's a default, it must be in the options list. - For `file_text` types: optional `acceptedExtensions` array (default: `[".txt", ".md", ".csv", ".json", ".html"]`). - `options` is only set when `type` is `select`. --- ## Step 3 — Create the artifact You'll generate an HTML file containing the workflow's specific UI, then call `mcp__cowork__create_artifact` with that file's path. **Sub-steps:** 1. If `mcp__cowork__create_artifact` isn't loaded, load it via ToolSearch. 2. Take the artifact template below and substitute the placeholders: - `__WORKFLOW_DATA_JSON__` → the JSON-stringified workflow object you synthesized in Step 2. Embed it directly as a JS object literal — do not wrap in a string. Preserve key names exactly as in the schema. - `__DISPLAY_NAME__` (in the `` tag) → the workflow's `displayName`, HTML-escaped. 3. Write the resulting HTML to `/tmp/workflow-<filename>.html`. 4. Call `create_artifact` with: - `id`: the workflow's `filename` - `description`: `<displayName>: <shortDescription>. Open from the sidebar to fill the form and copy the run-prompt to your clipboard.` - `html_path`: the absolute path to the file you wrote. ### Artifact template Use this verbatim (after substitution). Keep the structure and JS logic exactly as written — it handles clipboard fallback, form validation, and the run-prompt assembly: ```html <!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8" /> <title>__DISPLAY_NAME__

``` --- ## Step 4 — Confirm and explain how to run Tell the user: > Done. The **** workflow is in your Cowork sidebar artifacts menu. > > **To run it:** > 1. Open the artifact from the sidebar. > 2. Fill in the form fields. Defaults are pre-filled where I had them. > 3. Click **Run workflow** — the run-prompt is copied to your clipboard. > 4. Switch to a Cowork chat (any chat) and paste, then hit Enter. I'll execute the workflow with your inputs. > > **To edit it:** open the artifact and click **Edit workflow**. The edit-prompt is copied to your clipboard. Paste it into a chat and we'll iterate on the workflow's design and update the artifact in place. > > **For a portable copy:** click **Download as .md** at the bottom of the artifact. You'll get a `.md` file with the full workflow definition. We recommend keeping these in `Claude_Brain/WORKFLOWS/` for organization — you can use them with other LLMs (paste into ChatGPT or Gemini), version-control them in git, or share with teammates. Stop. Do not produce any further output unless the user asks.

Sign in to use workflows