Loading ToolBench
MCP server for intent compilation, uncertainty detection, and truth verification for AI agents
All 10 tools have descriptions and visible schemas, but quality is significantly below production baseline. Tool names lack clear action verbs (veritas_* prefix is domain-specific but not action-oriented). Schemas are present but contain numerous extraneous parameters (include_content, intent, since, token_budget, max_results, cursor) that appear copy-pasted across all tools and are never explained in descriptions. Parameter descriptions are minimal or absent. No output schemas documented. Error handling guidance is absent. The server conflates query/pagination parameters with core tool logic, violating single-responsibility principle.
Check internal consistency of multiple claims in text
Check uncertainty level in a claim with optional context
Compile natural language into a structured IntentSpec with domain, entities, constraints, and ambiguities
Detect ambiguities in a natural language intent and list options for each
Extract and classify individual claims from a body of text
Generate the single highest-impact clarification question for an ambiguous intent
Parse and analyze causal relationships in text
All tools share identical extraneous parameters (include_content, intent, since, token_budget, max_results, cursor) with no descriptions. These appear to be copy-pasted query/pagination boilerplate unrelated to core tool logic, violating single-responsibility principle and confusing LLM tool selection.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract structured data without knowing response field names and types. Production tools must document return types.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
Score confidence level in a claim or statement
Synthesize verified information from text by extracting, verifying, and checking consistency of claims
Verify a factual claim against known facts and rate its reliability
Tool names use domain prefix (veritas_*) but lack clear action verbs. Names like 'veritas_compile_intent' and 'veritas_detect_ambiguity' are domain-specific but do not follow verb_noun convention (compile_intent, detect_ambiguity). LLMs rely on action verbs to infer intent before reading descriptions.
Parameter descriptions are missing or trivial. 'include_content' (boolean, default false) has no description of what content it includes or when to set it. 'intent' enum with values ['exists','ids','summary','full'] has no explanation of what each mode returns. LLMs cannot reason about parameter selection without descriptions.
No error handling guidance. Tool descriptions do not explain failure modes, recovery steps, or when to retry. LLMs receive no actionable error context to guide next steps.