Loading ToolBench
A portable AI assistant engine with task extraction, memory management, and MCP server integration for Claude Desktop
Claude BeestMode presents a single tool 'extract_tasks_from_text' with severe definitional gaps. While the tool has a description and input schema present in the code, the implementation reveals critical issues: (1) The description is vague about what 'AI analysis' means and lacks guidance on when/why to use this tool vs alternatives; (2) Input parameters lack descriptions entirely, 'text' and 'source' parameters have type declarations but no descriptive text explaining what format or constraints apply to each; (3) No output schema is documented in the visible code, making it impossible for LLMs to understand what structured data they will receive; (4) The tool name 'extract_tasks_from_text' follows verb_noun convention but conflates two concerns: extraction AND enrichment ('with context enrichment from memory'). The description promises memory enrichment and constraint validation that are not represented as separate concerns or tool parameters, violating single-responsibility principle. (5) Error handling is not evident in the code sample provided, no guidance on recovery if text is empty, malformed, or if AI analysis fails. (6) The 'source' parameter expects specific enum values ('morning_pages', 'unknown') per the Input description, but this constraint is not documented in parameter descriptions and no guidance is given about valid enums.
Extract actionable tasks from morning pages and other text sources using AI analysis, with context enrichment from memory and validation against agent constraints
Input parameter descriptions are completely missing. Parameters 'text' and 'source' have types but no descriptions explaining expected format, constraints, or when to use each value. LLMs cannot infer whether 'text' should be plaintext, markdown, or structured; or what 'source' values beyond 'morning_pages' and 'unknown' are valid.
No output schema documented. The tool returns extracted tasks but the visible code provides no schema definition of what fields are in the response (e.g., is it a list of task objects? Does each task have name, priority, deadline, source_line? Are there validation results?). Without this, LLMs cannot chain results to downstream tools or extract required data.
Tool description is vague and conflates multiple concerns. It promises 'context enrichment from memory and validation against agent constraints' but these are not exposed as parameters or explained as behavior. This violates single-responsibility, is this a pure extraction tool, or does it also query memory and validate? The description should clarify what this tool does vs. what preprocessing users must do before calling it.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 38 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No error handling guidance. What should the LLM do if text is empty, null, or malformed? If AI analysis fails? If memory enrichment is unavailable? The code provides no documented error responses or recovery paths.
Parameter 'source' implies an enum (e.g., 'morning_pages', 'unknown') but the enum values are not declared in the schema and are only hinted at in the Input description. JSON Schema should use 'enum' constraint; the description should list valid values explicitly.