Loading ToolBench
Agent-Model Recommendation Advisor. Fetches available models from OpenCode subscriptions and OpenRouter benchmarks, reads agent selection criteria from the Gentle AI guide, and helps recommend the best model for each agent/sub-agent.
Three tools with clear naming (list_, get_) and detailed descriptions (194-250 chars each). All tools have input schemas with type definitions and descriptions. However, output schemas are not formally documented, responses are returned as JSON text without explicit schema declarations. Parameter descriptions are thorough but lack constraint details (e.g., no min/max for enrich boolean, no format guidance for query string). Error handling returns text-based messages but lacks recovery guidance or error classification. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all three being read-only operations.
Read the Gentle AI agent selection criteria from guia_gentle_ai.md. Each agent section defines: - What the agent does (description) - CRITERIOS DE SELECCIÓN: what qualities the model needs (context, reasoning, speed, cost, etc.) - Razonamiento para la selección: why certain tradeoffs matter Use this to understand WHAT each agent needs BEFORE picking a model. Pass an agent ID to get only that agent's criteria. AGENT GROUPS (display recommendations in this order): 1. Orchestrator: gentle-orchestrator 2. SDD agents: sdd-init, sdd-onboard, sdd-explore, sdd-propose, sdd-spec, sdd-design, sdd-tasks, sdd-apply, sdd-verify, sdd-archive 3. Review (4R): review-risk, review-readability, review-reliability, review-resilience, review-refuter 4. Judgment Day: jd-judge-a, jd-judge-b, jd-fix-agent
Search OpenRouter for detailed benchmarks and pricing for a specific model. Use this when you need deeper data on a particular model beyond what list_available_models returns. Returns: model ID, name, pricing, context length, and artificial_analysis benchmarks.
Fetch all available AI models from your OpenCode subscriptions (Go and/or Zen), cross-referenced with OpenRouter benchmarks when available. Returns for each model: - ID, name, provider - Which subscription(s) it belongs to (go, zen, or both) - Pricing per 1M tokens (input/output) — from OpenRouter - Context length (max tokens) - Benchmarks: intelligence_index, coding_index, agentic_index (Artificial Analysis) - Reasoning: available effort levels (e.g. ["xhigh","high"]) and default — when the model supports reasoning_effort Use this tool FIRST to understand what models are available before making recommendations.
Output schemas not formally documented. Tools return JSON text without explicit response schema declarations. LLMs cannot reliably parse or chain results.
No tool annotations despite all three tools being read-only. Missing readOnlyHint declarations prevent clients from optimizing caching and retry logic.
Error handling returns plain text messages without recovery guidance. 'Error: <message>' does not tell LLM what to do next or whether to retry.
Parameter 'query' in get_model_benchmarks lacks format guidance. Description says 'Model ID or name' but does not specify valid formats, length limits, or case sensitivity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
get_agent_criteria agent parameter filtering logic incomplete in visible code. Lines cut off mid-implementation; cannot verify full parameter validation or edge case handling.