Loading ToolBench
The Autonomous Nervous System for AI Agents — local-first MCP middleware with temporal graph memory, active coercion, and proactive goal alignment.
Single tool 'memento_compact' has a well-structured schema with proper types and enums, and a detailed description (240+ chars) explaining strategies and use cases. However, the tool lacks error handling guidance, output schema documentation, and parameter descriptions could be more prescriptive about constraints. The description mentions 'No LLM call' but doesn't explain when to use this tool vs alternatives or what happens on edge cases (empty text, invalid strategy). Parameter 'max_tokens' lacks range constraints (min/max). Overall solid foundation but missing production-grade error recovery and output documentation.
Compress natural-language text to save tokens while preserving technical substance (code, URLs, file paths). Strategies: caveman, truncate, graphify, ponytail. 'caveman' applies the MIT-licensed caveman-compress rule-based tersification; 'truncate' hard-cuts at the token budget; 'graphify'/'ponytail' are native token-light heuristics. No LLM call.
Output schema not documented. Tool returns TextContent with a formatted string, but LLMs cannot plan downstream operations without knowing the exact response structure (e.g., whether 'compacted' is a separate field or embedded in text).
Parameter 'max_tokens' lacks numeric constraints. No minimum or maximum specified; LLMs could pass 0, negative, or absurdly large values. Should declare range (e.g., 1 - 10000).
No error handling guidance. Tool silently defaults strategy to 'caveman' if invalid; silently defaults max_tokens to 60 if unparseable. LLMs cannot distinguish between intentional defaults and error recovery, risking silent misuse.
Parameter descriptions lack actionable detail. 'The text to compact' is minimal; should explain expected input size, supported formats (code, URLs, file paths), and what 'preserving technical substance' means in practice.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Tool name 'memento_compact' is domain-specific and does not follow verb_noun convention (e.g., 'compress_text' or 'compact_text' would be clearer). LLMs may not immediately recognize this as a text compression tool.