Loading ToolBench
OpenClaw-style adapter that routes proposed local actions through Neura Relay before execution, converting refs-only proposed actions into Action Cards and returning Decision Receipts before developer-owned execution.
Single tool with significant definition gaps. Tool name 'neura_relay_preflight_action' lacks a clear action verb (should be 'create_' or 'request_'). Description is 156 chars but vague about what 'refs-only' means and lacks WHEN/WHY guidance. Input schema present but parameters lack descriptions entirely, 'proposedAction', 'authority', 'evidenceRefs', 'ruleRefs', 'riskCategory' are all undescribed objects/arrays. No output schema documented. No error handling guidance. Tool appears domain-specific (governance/decision receipts) but definitions assume deep prior knowledge of Neura Relay concepts.
Create a refs-only Action Card and request a Neura Decision Receipt before a local agent action executes.
Tool name lacks action verb. 'neura_relay_preflight_action' is a noun phrase; should start with 'create_', 'request_', or 'submit_' to signal intent to LLM.
All input parameters lack descriptions. 'proposedAction', 'authority', 'evidenceRefs', 'ruleRefs', 'riskCategory' are defined as objects/arrays with no explanation of structure, expected content, or format. LLM cannot infer what to pass.
No output schema documented. Tool description does not specify what 'Decision Receipt' contains, what fields are returned, or how the LLM should interpret the response.
Description assumes domain knowledge ('refs-only Action Card', 'Neura Decision Receipt'). No explanation of what these terms mean or when to use this tool vs alternatives.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | 2026-07-28+ | v2 |
No error handling guidance. Tool description does not explain failure modes, recovery steps, or what to do if preflight review is rejected.