Loading ToolBench
MCP local de découverte et sélection déterministe de skills validés
Server defines 5 tools with explicit JSON schemas and descriptions in French. All tools have proper input schemas with type constraints, enums, and maxLength/minItems/maxItems bounds. Naming follows verb_noun pattern (detect, list, get, plan, refresh). However, descriptions are terse (10-20 chars in French), lack actionable context for LLM selection, and parameter descriptions are minimal. No output schemas documented. Error handling is absent from tool definitions. Tools are read-only (good security posture) but lack guidance on when to call each vs. alternatives.
Detecte localement le profil borne d'une cible. N'execute aucun fichier.
Retourne le contenu borne d'un skill valide par identifiant. Aucun chemin libre n'est accepte.
Decouvre et liste uniquement les packages de skills valides.
Detecte une cible, charge le registre valide et rend un plan deterministe. N'execute aucun skill.
Reconstruit explicitement un registre depuis les racines autorisees.
Descriptions are under 20 characters and lack LLM-actionable context. E.g., 'Detecte localement le profil borne d'une cible' (French, ~45 chars) does not explain WHEN to call this vs. phases_agents_plan, what it returns, or prerequisites.
No output schemas documented. LLMs cannot infer what fields to expect from detect, list_skills, get_skill, plan, or refresh_skills. This blocks downstream tool chaining and forces agents to guess field names.
Parameter descriptions are missing or minimal. E.g., 'root_ids' and 'today' in list_skills lack explanation of their purpose, format, or constraints beyond JSON Schema. LLMs cannot reason about when to populate these.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
No error handling guidance in tool definitions. No indication of retryable vs. fatal errors, or what the LLM should do if a skill is not found, a root_id is invalid, or the target path is inaccessible.
Tool composition unclear. Relationship between detect → list_skills → get_skill → plan is not documented. LLMs cannot determine the intended workflow or when to call each tool in sequence.