Loading ToolBench
An agent server for verification case management with document extraction, rule evaluation, release preparation, and human signature workflows
VeriRemit Agent defines 4 tools with clear verb-noun naming (get_, extract_, prepare_, request_) and explicit JSON Schema input parameters. All tools have descriptions (50-90 chars) and properly typed caseId parameters with minLength constraints. However, output schemas are not documented, the code shows summarizeCase() returns an object but the schema structure is not formally declared. Parameter descriptions are minimal (only 'The unique identifier...'). No error handling guidance, no recovery hints, and no documentation of timeouts or side effects visible in tool definitions. Tool composition is sound (each does one thing), but descriptions lack context about prerequisites, when to use each tool, or what happens on failure.
Start document extraction and deterministic verification for a newly created case. This cannot approve exceptions or sign anything.
Read sanitized case status, deterministic rule outcomes, human-review state, and release/signature state.
Request release-packet preparation through the authoritative server gate. This cannot sign or authorize payment.
Create the gated Foxit eSign envelope for the configured human signer after release preparation. This requests a human signature; it cannot sign, approve verification, or authorize payment.
Output schemas not documented. Code shows summarizeCase() returns {found, id, status, rules[], humanVerification, releasePrepared, signatureStatus} but this structure is not formally declared in tool definitions. LLMs cannot plan downstream calls or extract fields without knowing the response shape.
Parameter descriptions are minimal (only 'The unique identifier of the verification case'). No guidance on format, constraints, or when to use each tool. Descriptions should explain WHAT the tool does, WHEN to call it, and any prerequisites (e.g., 'extract_case requires a newly created case; call get_case_summary first to check status').
No error handling guidance. Tool descriptions do not explain what errors can occur, whether they are retryable, or what the LLM should do next. E.g., extract_case has a 60s timeout but no guidance on timeout recovery.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Tool descriptions lack state-change clarity. extract_case, prepare_release, and request_human_signature are WRITE operations but descriptions do not explicitly state 'This modifies the case state' or 'This is irreversible'. Agents need to know which calls are safe to retry.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). get_case_summary should be marked read-only; extract_case, prepare_release, request_human_signature should be marked destructive or have idempotency guidance.