Loading ToolBench
🛡️ Make AI interrupt itself and challenge your thinking. Turns Claude into a philosophical sparring partner who actively contradicts comfortable conversations and challenges confirmation bias.
Four tools with complete schemas and detailed descriptions, but significant gaps in naming clarity, parameter validation guidance, and error handling. Tool names lack action verbs (ethics_check, ethics_guide, ethics_learn, critical_thinking are descriptive but not verb-noun patterns). Descriptions are verbose (200-400+ chars) and include examples that could be reused literally by LLMs. Parameters are well-typed with enums where appropriate (category in ethics_learn), but lack constraint documentation (e.g., no min/max for string lengths, no regex patterns stated). Output schemas are not documented in the source. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Security: sessionId and previousConcerns parameters suggest stateful tracking without documented storage/privacy model.
🧠 CONFIRMATION BIAS & CRITICAL THINKING ANALYSIS - Call this tool to analyze AI responses for confirmation bias and intellectual rigor. USE WHEN: (1) An AI response seems overly agreeable or one-sided, (2) You suspect confirmation bias in an AI's reasoning, (3) Need to evaluate the critical thinking quality of an AI response, (4) Want to identify missing perspectives or counterarguments, (5) Analyzing responses to controversial or complex topics. REQUIRED: 'aiResponse' (the AI response to analyze) and 'userRequest' (the original request that prompted it). OPTIONAL: 'context' for additional background, 'sessionId' for bias pattern tracking.
🛡️ COMPREHENSIVE ETHICAL ANALYSIS - Call this tool when you need to analyze AI conversations or responses for ethical concerns. USE WHEN: (1) Reviewing potentially sensitive conversations, (2) Analyzing AI responses that might contain bias or harmful content, (3) Checking for confirmation bias in AI outputs, (4) Evaluating conversations involving personal data, controversial topics, or vulnerable populations. REQUIRED: Both 'conversation' (the AI response/conversation to analyze) and 'userRequest' (the original user request that prompted it). OPTIONAL: Add 'context' for situational details, 'focusAreas' array for specific ethical areas to emphasize (e.g., ['privacy', 'bias', 'confirmation bias']), 'previousConcerns' for ongoing ethical issues, 'sessionId' for tracking patterns.
🧭 PROACTIVE ETHICAL GUIDANCE - Call this tool when you need ethical advice for complex scenarios or dilemmas. USE WHEN: (1) Facing ethical dilemmas or moral questions, (2) Need guidance on best practices for sensitive situations, (3) Planning actions that could have ethical implications, (4) Seeking framework-based ethical analysis, (5) Need help navigating stakeholder conflicts or competing values. REQUIRED: 'scenario' (describe the ethical situation or dilemma). OPTIONAL: 'domain' for context (healthcare, technology, business, etc.), 'stakeholders' array listing affected parties, 'sessionId' for tracking guidance patterns.
Tool names lack action verbs. 'ethics_check', 'ethics_guide', 'ethics_learn', 'critical_thinking' are descriptive but do not follow verb_noun pattern (e.g., 'analyze_ethics', 'get_ethical_guidance', 'log_ethical_concern', 'analyze_bias'). LLMs rely on verb-first names to infer intent before reading descriptions.
Descriptions include example values ('privacy', 'bias', 'confirmation bias', 'healthcare setting') that LLMs may reuse literally rather than adapt to context. Replace examples with enum constraints or formal patterns.
No output schema documented. Tool descriptions do not specify what fields are returned, their types, or structure. LLMs cannot plan downstream calls or extract data without knowing the response shape.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
📚 ETHICAL CONCERN LOGGING - Call this tool to log and learn from identified ethical concerns. USE WHEN: (1) You've identified a specific ethical concern that should be recorded, (2) Building pattern recognition for future ethical oversight, (3) Contributing to institutional knowledge about ethical issues, (4) After using ethics_check and want to log specific concerns found. REQUIRED: 'concern' (description of the issue), 'category' (must be one of the predefined categories), 'severity' (low/medium/high/critical), 'recommendation' (how to address it). OPTIONAL: 'sessionId' for pattern tracking.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. A failed ethics analysis returns nothing actionable.
sessionId and previousConcerns parameters suggest stateful session tracking, but no documentation of storage mechanism, privacy guarantees, or data retention policy. Agents cannot reason about side effects or compliance implications.