Retrieval Playbook for AI

Retrieval Playbook for AI

Overview

Playbook สำหรับตัดสินใจว่าจะ retrieve context อะไรให้ AI - เลือกแค่ที่จำเป็น ตัดที่ไม่เกี่ยว จัดลำดับให้เหมาะสม

Why This Matters

• Relevance: เฉพาะที่เกี่ยวข้อง
• Token efficiency: ไม่เสีย token กับข้อมูลไม่เกี่ยว
• Better results: AI focus ได้ดีขึ้น
• Faster: น้อย context = เร็วขึ้น

Decision Tree

Task received
    ↓
Is it code-related?
    ├─ Yes → Retrieve code files
    └─ No → Skip code
         ↓
Need current state?
    ├─ Yes → Include recent changes
    └─ No → Skip history
         ↓
Need examples?
    ├─ Yes → Include 1-2 examples
    └─ No → Skip examples
         ↓
Assemble context

Retrieval Rules

Rule 1: Only Direct Dependencies

❌ Include entire codebase
✅ Include only files directly related

Example:
Task: Fix bug in auth.ts
Include:
- auth.ts (the file)
- types.ts (if auth.ts imports it)
- config.ts (if auth.ts uses it)

Don't include:
- unrelated files
- test files (unless debugging tests)
- documentation

Rule 2: Snippets Over Full Files

❌ Include entire 500-line file
✅ Include 20-line relevant function

Example:
Task: Fix validateToken function
Include:
- validateToken function (lines 45-60)
- Related types (5 lines)
- Helper functions used (10 lines)

Total: ~35 lines vs 500 lines
Savings: 93%

Rule 3: Recent Over Old

✅ Last 3 commits
❌ Full git history

✅ Current implementation
❌ Deprecated code

Rule 4: Examples Only When Needed

Include examples when:
✓ New pattern/concept
✓ Complex logic
✓ User explicitly asks

Skip examples when:
✗ Simple CRUD
✗ Standard patterns
✗ Self-explanatory

Context Prioritization

Priority 1: Critical (Always Include)

- File with the bug/feature
- Error messages
- Relevant types/interfaces
- Direct dependencies

Priority 2: Important (Include if Space)

- Related functions
- Configuration
- Recent changes
- Test cases

Priority 3: Nice-to-Have (Usually Skip)

- Full documentation
- Examples
- Comments
- Historical context

Retrieval Strategies by Task Type

Bug Fix

Retrieve:
✓ File with bug (snippet)
✓ Error message
✓ Stack trace
✓ Related types

Skip:
✗ Entire file
✗ Unrelated files
✗ Documentation
✗ Tests (unless test is failing)

Feature Development

Retrieve:
✓ Similar existing features (1 example)
✓ Relevant types
✓ API contracts
✓ Configuration

Skip:
✗ Multiple examples
✗ Full documentation
✗ Unrelated features

Code Review

Retrieve:
✓ Changed files (diff)
✓ Related files (if imports changed)
✓ Style guide (summary)

Skip:
✗ Unchanged files
✗ Full style guide
✗ Historical context

Refactoring

Retrieve:
✓ File to refactor
✓ Files that import it
✓ Tests for it

Skip:
✗ Unrelated files
✗ Documentation
✗ Examples

Smart Filtering

Filter by Recency

// Get files changed in last 7 days
const recentFiles = gitLog()
  .filter(commit => commit.date > sevenDaysAgo)
  .flatMap(commit => commit.files);

Filter by Relevance

// Score files by relevance
const relevantFiles = allFiles
  .map(file => ({
    file,
    score: calculateRelevance(file, task)
  }))
  .filter(f => f.score > 0.7)
  .sort((a, b) => b.score - a.score)
  .slice(0, 5);  // Top 5 only

Filter by Size

// Prefer smaller, focused files
const files = candidates
  .filter(f => f.lines < 200)  // Skip large files
  .sort((a, b) => a.lines - b.lines);

Context Assembly

Order Matters

1. Summary (what we're doing)
2. Error/issue (if applicable)
3. Relevant code (snippets)
4. Types/interfaces
5. Configuration (if needed)

❌ Don't:
- Start with full files
- Mix unrelated contexts
- Bury important info

Example Assembly

# Task: Fix login bug

## Issue
Login fails with "Invalid token" error

## Code
```typescript
// auth.ts:45-60
function validateToken(token: string) {
  // Bug: missing null check
  return jwt.verify(token, SECRET);
}

Types

interface Token {
  userId: string;
  exp: number;
}

Goal

Add null check before jwt.verify

---

## Metrics to Track

### Retrieval Efficiency
```typescript
interface RetrievalMetrics {
  filesRetrieved: number;
  tokensUsed: number;
  relevanceScore: number;  // 0-1
  taskSuccess: boolean;
}

// Good retrieval:
{
  filesRetrieved: 3,
  tokensUsed: 500,
  relevanceScore: 0.9,
  taskSuccess: true
}

// Bad retrieval:
{
  filesRetrieved: 20,
  tokensUsed: 5000,
  relevanceScore: 0.3,
  taskSuccess: false
}

Anti-Patterns

❌ Kitchen Sink Approach

"Include everything just in case"
Result: 10,000 tokens, AI confused

❌ No Filtering

"Retrieve all files that mention 'user'"
Result: 50 files, mostly irrelevant

❌ Full File Dumps

"Here's the entire codebase"
Result: Context limit exceeded

❌ No Prioritization

"All files are equally important"
Result: Critical info buried

Quick Checklist

Before retrieving, ask:
☐ Is this directly related to the task?
☐ Can I use a snippet instead of full file?
☐ Is this the most recent version?
☐ Will AI actually use this?
☐ Am I under my token budget?

If any answer is "no", reconsider including it.

Summary

Retrieval Playbook: เลือก context อย่างชาญฉลาด

Rules:

1. Only direct dependencies
2. Snippets over full files
3. Recent over old
4. Examples only when needed

Priorities:

• P1: Critical (always)
• P2: Important (if space)
• P3: Nice-to-have (skip)

By Task:

• Bug fix: Error + snippet
• Feature: Example + types
• Review: Diff only
• Refactor: File + dependents

Target:

• 3-5 files max
• 500-1000 tokens
• 90%+ relevance