Loading ToolBench
A type-safe Python SDK for executing Airbyte connectors from an application or agent. Exposes 50+ third-party APIs (Stripe, Zendesk, HubSpot, etc.) as typed Python objects or generic hosted executors.
Three tools with partial schemas and descriptions. inspect_connector and read_skill_docs have empty input schemas ({}), violating schema requirements. execute has a structured schema with typed parameters. Descriptions exist but are verbose and lack clarity on when to use each tool vs. alternatives. No parameter descriptions for execute's 'params' object. Missing error handling guidance, output schemas, and pagination details. Tool naming follows verb_noun pattern (inspect_, read_, execute) but descriptions are procedural rather than declarative.
Execute a connector operation. Before your first execute call, call inspect_connector(), then read_skill_docs(), then read_skill_docs(section=<exact section id>) for the entity/action you plan to execute. Do not guess entity, action, or parameter names. Exact names, params, fields, relationships, and examples live in read_skill_docs. Use entity and action names exactly as documented by the schema and docs. For collection responses, keep outputs small with filters and small limits. If output is too large, refine the query instead of repeating the same broad call.
Inspect the connector's hosted metadata and resolve its docs skill id. Local/offline connectors (no docs provider) get a local-mode payload with a warning instead of a hosted inspection.
Read the connector's usage docs, rendered to text. Omit section for the outline and general guidance; pass an exact section id for full details. Local/offline connectors return the full generated docs and ignore section.
inspect_connector and read_skill_docs have empty input schemas ({}). inspect_connector accepts no parameters but read_skill_docs accepts 'section', schema must document this.
execute's 'params' parameter is typed as 'object' with no schema constraints, description, or examples. LLMs cannot infer valid parameter names or structure. Must document entity/action pairs and their required params.
Descriptions are procedural (tell agent to call inspect first, then read_skill_docs, then execute) rather than declarative (what does this tool do?). Descriptions exceed 1024 chars and bury key details. Violates pattern:tool-description baseline of 10 - 1024 chars.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields (docs_skill_id, section IDs, entity/action names). Violates pattern:tool baseline.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
No error handling guidance. If execute fails (invalid entity, missing required param, API error), LLM has no recovery path. Missing pattern:recovery-guide and pattern:error-classification.