Feature: Skills/Tools Prompt Budget Management

[Sahil5963]

Summary

Implement configurable budget limits for skills and tool definitions in prompts to prevent context overflow.

Problem

When loading many skills or tools:

• Skill descriptions can consume excessive context
• No limit on how much space tool definitions take
• Large skill files can blow up the prompt
• Users have no visibility into token usage

Proposed Solution

Configuration Options

interface SkillsBudgetConfig {
  enabled?: boolean;                    // Default: false
  
  limits?: {
    maxSkillsInPrompt?: number;         // Default: 150
    maxSkillsPromptChars?: number;      // Default: 30_000 (30K chars)
    maxSkillFileBytes?: number;         // Default: 256_000 (256KB per file)
    maxToolDefinitionChars?: number;    // Default: 50_000
  };
  
  optimization?: {
    compactPaths?: boolean;             // Replace /Users/x/ with ~/
    minifyDescriptions?: boolean;       // Shorten verbose descriptions
    prioritization?: 'recent' | 'frequent' | 'alphabetical';
  };
  
  warnings?: {
    enabled?: boolean;                  // Warn when approaching limits
    thresholdPercent?: number;          // Warn at 80% usage
  };
}

// Usage
const chat = createChatWithTools({
  skillsBudget: {
    enabled: true,
    limits: {
      maxSkillsInPrompt: 100,
      maxSkillsPromptChars: 25_000,
    },
    optimization: {
      compactPaths: true,
      prioritization: 'frequent',
    },
    warnings: {
      enabled: true,
      thresholdPercent: 80,
    }
  }
});

Smart Fitting Algorithm

// Binary search to fit maximum skills within budget
function fitSkillsToBudget(skills: Skill[], config: SkillsBudgetConfig): Skill[] {
  const { maxSkillsInPrompt, maxSkillsPromptChars } = config.limits;
  
  // 1. Limit by count
  let candidates = skills.slice(0, maxSkillsInPrompt);
  
  // 2. Binary search to fit char budget
  let lo = 0, hi = candidates.length;
  while (lo < hi) {
    const mid = Math.ceil((lo + hi) / 2);
    const prompt = formatSkills(candidates.slice(0, mid));
    if (prompt.length <= maxSkillsPromptChars) lo = mid;
    else hi = mid - 1;
  }
  
  return candidates.slice(0, lo);
}

Path Compaction (Token Savings)

// Before: /Users/alice/.config/copilot/skills/github/SKILL.md
// After:  ~/.config/copilot/skills/github/SKILL.md
// Saves ~5-6 tokens per skill path

optimization: {
  compactPaths: true,  // Enable path compaction
}

Use Cases

• Large skill libraries with 100+ skills
• Enterprise deployments with many custom tools
• Multi-tenant systems with per-user skills
• Resource-constrained environments

Benefits

• ✅ Fully optional - disabled by default
• ✅ Configurable limits - tune for your context window
• ✅ Prevents prompt overflow from too many skills
• ✅ Smart fitting maximizes skills within budget
• ✅ Path compaction saves tokens
• ✅ Warnings before hitting limits
• ✅ Prioritization keeps most useful skills

[Sahil5963]

Consolidated into #72 (Intelligent Tool Management System) - see Result Truncation & Budget section

[Sahil5963]

Reopening - this is a distinct feature from #72. Budget Management focuses on context SIZE limits and token tracking, while #72 focuses on tool discovery. Budget management is essential infrastructure regardless of search implementation.