Local-first form engine for LLM agents — Typeform-quality forms generated on-the-fly
Three tools with clear naming (verb_noun pattern) and reasonable descriptions (100-200 chars). All tools have input schemas with typed properties. However, parameter descriptions are sparse or missing entirely, most params lack guidance on format, constraints, or valid ranges. Output schemas are undocumented; LLMs cannot predict response structure. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Composition is sound (each tool has one responsibility), but missing idempotency guarantees and pagination support for list_submissions.
Create a formloop form from a JSON spec. Returns the form URL for the user to fill. The spec must include a `title` (string) and `blocks` (array of question objects). Question kinds: mc (multiple choice), multi (checkboxes), text, textarea, yn (yes/no), number, scale (1-N), html-pick (clickable HTML cards), rank (drag-and-drop ordering), scale-preview (slider with live HTML preview), html (context-only, no answer). Each question block needs `kind`, `id`, and `title`. Options are required for mc, multi, html-pick, rank.
List all submissions for a persistent form. Only works with forms created with persistent: true.
Poll until an ephemeral form is submitted, then return the answers. Use after create_form to wait for the user to fill the form. Returns the structured answers keyed by question label.
Parameter descriptions missing or incomplete. 'spec' param in create_form lacks guidance on block structure, required fields per question kind, or validation rules. 'timeout_seconds' in wait_for_response has no range (min/max). LLMs cannot infer valid input without explicit constraints.
Output schemas undocumented. create_form returns 'form URL' but structure is unspecified. wait_for_response returns 'structured answers', what fields? What types? list_submissions returns submissions, what fields per submission? LLMs cannot plan downstream calls without knowing response shape.
No error handling guidance. Tools lack recovery hints (e.g., 'If form_id is invalid, call create_form again'). No categorization of errors as retryable vs fatal. Agents cannot self-correct on failure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
list_submissions lacks pagination. No limit, offset, or cursor parameters. If a persistent form has hundreds of submissions, returning all at once wastes tokens and risks context window exhaustion. Should accept limit (default 20) and offset/cursor.
Idempotency not guaranteed. create_form does not document whether repeated calls with identical spec produce the same form_id or new forms. wait_for_response polling behavior on timeout is unclear, does it return partial data or error? Agents need idempotency guarantees for safe retries.