Loading ToolBench
Woven turns one request in ChatGPT or Codex into a complete compatible cart and one exact confirmation. Composes retail shopping carts under hard requirements, compatibility, budget, availability, and Singapore pickup constraints using deterministic camping engine or bounded open-world research workflow.
Woven demonstrates solid tool definition fundamentals with 14 well-named, verb-prefixed tools and comprehensive input schemas using Zod validation. All tools have descriptions (avg ~150 chars) and typed parameters with constraints. However, output schemas are undocumented, responses return MissionView objects without published field definitions, forcing LLMs to infer structure. Error handling is present but generic (DomainError wrapper). Tool composition is strong: start_mission → build_carts → select_cart → swap_cart_item → checkout flow chains logically. Security concerns: admin-only tools (set_scenario, set_offer_price) lack explicit permission gates in visible code. Parameter descriptions are adequate but lack dependency hints (e.g., 'select_cart requires a valid missionId from start_mission'). No pagination support visible despite potential for large result sets.
Refresh ranked carts for an existing Woven mission after price or inventory changes.
Confirm an exact checkout mandate, authorize payment via simulated Visa, and create an order. App-only.
Require a verified demo identity, revalidate a selected cart, and create an expiring checkout mandate. App-only.
Get the current demo scenario (normal, stockout, price-change, auth-decline, order-fail). Admin-only.
List all merchant-approved item alternatives. Admin-only.
Select one ranked cart in the Woven widget. App-only.
Enable or disable a merchant-approved alternative. Admin-only.
Output schemas undocumented. Tools return MissionView objects without published field definitions. LLMs cannot infer what fields to extract or how to chain downstream calls.
Admin-only tools (set_scenario, set_offer_price, set_offer_stock, set_alternative_active) lack explicit permission gate documentation. No visible scope declarations (e.g., 'admin:write') or permission check guidance in descriptions.
Parameter descriptions lack dependency hints. E.g., select_cart requires missionId from start_mission, but this prerequisite is not stated. Forces LLMs to infer call ordering.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2025-06-18+ | v2 |
Update an offer's price in cents. Admin-only.
Update an offer's stock quantity. Admin-only.
Set the demo scenario to simulate different failure modes. Admin-only.
Start Woven's simulated, server-enforced identity handoff before checkout. App-only.
Use when a user asks Woven to compose a complete retail cart under hard requirements, compatibility, budget, availability, and Singapore pickup constraints. Camping uses the deterministic engine; other categories use the bounded open-world research workflow.
Replace one item with an active merchant-approved alternative while preserving compatibility, pickup location, and budget. App-only.
Verify a signed receipt after order confirmation. App-only.
Error handling returns generic DomainError codes (INVALID_INPUT, INTERNAL_ERROR) without recovery guidance. E.g., 'INVALID_INPUT' does not tell LLM what to try next or which parameter failed.
No pagination support visible. list_merchant_alternatives and similar discovery tools may return unbounded results, risking context window exhaustion.