LLM Council Context Skill

Overview

Provide concise, high-signal context for consult_llm_council requests. This skill focuses on context budgeting, structuring information, and avoiding truncation or irrelevant details.

Core Guidance

• Prefer a short, structured summary over raw dumps.
• Lead with the goal and constraints.
• Include only the most relevant excerpts.
• Reserve budget for the model response.

Model Selection Flow

If the user requests a specific model (e.g., "use Claude" or "use Anthropic"), translate that intent into the models array for consult_llm_council.

When model names are uncertain, use this flow:

1. Call list_models to retrieve the exact available model names.
2. Pass one or more returned names to consult_llm_council via the models parameter.

This avoids guessing and ensures the server accepts the model selection.

Recommended Brief

Use a concise schema for repeatable structure:

typescript
1interface ContextBrief {
2  goal: string;
3  constraints: string[];
4  relevant_facts: string[];
5  avoid: string[];
6}

When To Load The Reference Guide

Open references/CONTEXT_GUIDE.md when:

• The user asks how to share context or what to include
• You need concrete examples (code review, PR review, architecture decision, bug fix, feature request)
• You need truncation/budget examples or provider context limits

How To Use This Skill

1. Gather the user’s goal, constraints, and the minimal relevant facts.
2. Propose a concise ContextBrief if none is provided.
3. Suggest what to omit or summarize to stay within budget.
4. If the context is large, include a short summary + targeted excerpts.

Resources

• references/CONTEXT_GUIDE.md: Full context-sharing guide with budgets, truncation examples, and scenario templates.