Multi-agent system for asynchronous communication between UI and backend using LLM-powered agents (Router, Planner, Coder, Tester, Chatter) for code generation and explanation
X-Coder exhibits severe definition quality gaps across all six tools. Tool names lack clear action verbs (chat_endpoint, route_request are vague; generate_response conflates multiple concerns). Descriptions are present but generic and lack LLM-optimization guidance. Input schemas are minimally specified, most parameters lack type constraints, enums, or validation rules. Output schemas are entirely undocumented. No error handling guidance, no parameter dependencies documented, no idempotency hints. The codebase shows internal agent orchestration but exposes a thin MCP interface with insufficient metadata for reliable agent reasoning.
Processes user queries through a multi-agent workflow (routing, planning, coding, testing, chat) and streams results back as Server-Sent Events
Creates a detailed plan for solving the user's request using the planner LLM agent
Generates code with explanations based on instructions, parsing output into code blocks and explanatory text
Generates contextual response from chatter agent, adapting role based on whether code was generated (presentation vs mentor role)
Routes user query to appropriate agent (PLANNER, CODER, or CHAT) using grammar-constrained LLM generation
Validates user input query for gibberish content using basic pattern matching and entropy analysis
No output schemas documented. Tools return unstructured responses (e.g., generate_code returns 'code blocks and explanatory text' with no field definitions). LLMs cannot plan downstream calls or extract structured data.
Input schemas lack type constraints and enums. 'query' and 'user_query' are bare strings with no format, length, or validation rules. No guidance on what constitutes valid input.
Tool names lack clear action verbs. 'chat_endpoint' and 'route_request' are vague; 'generate_response' conflates role-switching logic with response generation. Names should start with verb_noun (e.g., 'route_query', 'generate_plan_response').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
No error handling guidance. Descriptions do not explain failure modes, recovery steps, or when to retry. LLMs receive no actionable error context.
Parameter descriptions are minimal (10-50 chars). 'context' parameter in generate_response is described as 'Context dictionary' with no explanation of required fields, structure, or dependencies.