So your forms stop producing tasks with eleven empty columns.

Your ClickUp dashboard may depend on a custom field that nobody fills in. An empty field seldom raises an error. The task simply disappears from the view filtered by that field, and the team starts making decisions from an incomplete picture.

This build watches one ClickUp List, reads each new task’s title and description, asks Claude to classify the task against your existing field definitions, then writes only well-supported values back to that same task. It starts in a log-only mode so you can check its decisions before it touches live data.

What this actually does

A form creates a task called “Review vendor onboarding,” with the description “Finance needs to approve the Acme renewal before October 14.” Your field spec says that “Finance” can justify Workstream = Finance and “approve” can justify Work type = Approval. Claude returns each proposed value with the exact source phrase and a confidence level. Make checks that the phrase really occurs in the task text, maps the approved dropdown names to their actual ClickUp option IDs, and fills only fields that are still blank.

This classifies an existing task into a taxonomy your team already uses. It does not create subtasks or decompose vague work into new items. It also does not make a weak description more informative.

What this does not do: It does not know which department owns a vendor when that fact is absent from the title and description. It does not infer a budget, deadline, or priority from background knowledge. It does not overwrite a human value. It does not repair an inconsistent taxonomy or guarantee perfect dashboards. Tasks with insufficient evidence remain visibly incomplete for a person to review.

The architecture uses the Anthropic API through Make’s HTTP module. Make holds the connection and executes the writes unattended. A chat connector that asks a person to approve each action would break this particular workflow. A ClickUp webhook or Make’s ClickUp Watch Tasks module supplies the trigger; the model only proposes values and never receives a ClickUp write credential.

Cost illustration: Claude Haiku 4.5 is listed at $1 per million input tokens and $5 per million output tokens. If a task uses 1,500 input tokens and 350 output tokens, its model charge is about $0.00325. At 1,000 processed tasks per month, that is about $3.25 in Anthropic model charges, before Make subscription, Make credits, failed runs, retries, or unusually long descriptions. These token counts are budgeting assumptions, not measurements from this scenario. Check your actual usage after the dry run. Anthropic model pricing.

Phase 0 — Preconditions and field audit

Start with a ClickUp Unlimited plan or higher. ClickUp’s Free Forever plan has 60 Custom Field uses, where adding a value to a field on a task counts as a use. This tutorial assumes repeated automated writes, so Free Forever is unsuitable. Confirm your own plan and permissions before building. ClickUp Custom Field uses.

Create a Make scenario and obtain a ClickUp personal API token from an account permitted to read the target List and edit its fields. Create an Anthropic API key and store both credentials in protected Make connections or credentials. Do not put keys in task descriptions, prompts, audit logs, or a shared Notion page. Record the workspace ID, target Space ID, and target List ID; the List ID controls the narrow scope of this build. ClickUp authentication.

In Make, use ClickUp > List All Accessible Custom Fields for the target List, or use ClickUp’s Get List Custom Fields endpoint, GET /api/v2/list/{list_id}/field. Save a working table of each candidate field’s name, ID, type, and allowed values. For dropdowns, copy the exact type_config.options names and IDs. Check that the fields apply to the tasks and custom task types you intend to process. A field that appears elsewhere in the workspace may not be applicable here. Make ClickUp modules; ClickUp List Custom Fields.

The percentages above are illustrative. Calculate yours from existing tasks: tasks with a value in this field ÷ tasks examined × 100. Export or paginate through a representative set of tasks in the target List, and count each field separately. Use a defined period, such as the last 90 days, and record the sample size. If “External reference” is already 94% populated, leave it out. A field at 8% may be useful to automate, provided the descriptions contain enough evidence to fill it accurately. The coverage gate prevents building an elaborate scenario for a field your team already maintains.

Critical distinction: A dropdown write needs its existing option ID, not the visible label “Finance.” ClickUp documents { "value": "option_1_id" } for dropdowns. Test one harmless field on a test task and read it back. Do not assume a successful Make module run means the intended label appeared. ClickUp Custom Fields.

Phase 1 — Write the field spec

Spend ten minutes defining the small taxonomy the model is allowed to use. For each selected field, write down its meaning, allowed values, evidence required for each value, and the fallback. Keep this spec as a named version, such as field-spec-v1, beside the scenario. The model must not invent new dropdown choices.