context-optimization

Goal: Manage context window usage efficiently to maintain productivity in long sessions.

| Step | Action | Key Notes |

|------|--------|-----------|

| 1 | Writing | Save critical findings to persistent memory |

| 2 | Selecting | Load relevant memories at session start |

| 3 | Compressing | Create context anchors every 10 operations |

| 4 | Isolating | Delegate specialized tasks to sub-agents |

Key Principles:

• Use offset/limit or grep before reading large files -- never read entire files unnecessarily
• Combine search patterns with OR (\|) instead of sequential searches
• Thresholds: 50K consider compression, 100K required, 150K critical

---

1. Writing (Save Important Context)

Save critical findings to persistent memory via memory-management skill:

• Discovered architectural patterns
• Important business rules, cross-service dependencies
• Solution decisions

2. Selecting (Retrieve Relevant Context)

Load relevant memories at session start:

```javascript

mcp__memory__search_nodes({ query: 'relevant keywords' });

mcp__memory__open_nodes({ names: ['EntityName'] });

```

3. Compressing (Summarize Long Trajectories)

Create context anchors every 10 operations:

```markdown

=== CONTEXT ANCHOR [N] ===

Task: [Original task]

Completed: [Done items]

Remaining: [Todo items]

Findings: [Key discoveries]

Next: [Specific next step]

Confidence: [High/Medium/Low]

===========================

```

4. Isolating (Use Sub-Agents)

Delegate specialized tasks: broad exploration, independent research, parallel investigations.

---

File Reading

```javascript

// BAD: Reading entire files

Read({ file_path: 'large-file.cs' });

// GOOD: Read specific sections

Read({ file_path: 'large-file.cs', offset: 100, limit: 50 });

// GOOD: Use grep to find content first

Grep({ pattern: 'class SaveEmployeeCommand', path: 'src/' });

```

Search Optimization

```javascript

// BAD: Multiple sequential searches

Grep({ pattern: 'CreateAsync' }); Grep({ pattern: 'UpdateAsync' });

// GOOD: Combined pattern

Grep({ pattern: 'CreateAsync|UpdateAsync|DeleteAsync', output_mode: 'files_with_matches' });

```

Parallel Operations

```javascript

// GOOD: Parallel reads for independent files

[Read({ file_path: 'file1.cs' }), Read({ file_path: 'file2.cs' })];

```

---

| Anti-Pattern | Better Approach |

| -------------------------- | ------------------------------ |

| Reading entire large files | Use offset/limit or grep first |

| Sequential searches | Combine with OR patterns |

| Repeating same searches | Cache results in memory |

| No context anchors | Write anchor every 10 ops |

| Not using sub-agents | Isolate exploration tasks |

| Forgetting discoveries | Save to memory entities |

---

Token Estimation: 1 line ~ 10-15 tokens | 1 page ~ 500 tokens | Avg file ~ 1-3K tokens

Thresholds: 50K: consider compression | 100K: required | 150K: critical - save & summarize

• Always plan and break many small todo tasks
• Always add a final review todo task to review the works done at the end to find any fix or enhancement needed