Agent 'Forgets' Instructions or Hallucinates Libraries

<h3>Step 1: Use .antigravityignore</h3>
<p>Just like `.gitignore`, you can tell the AI to ignore specific directories and files that pollute the context.</p>
<pre><code># .antigravityignore
# Legacy code and backups
legacy/
v1_backup/
deprecated/
old_*

# Build outputs and dependencies
node_modules/
dist/
build/
target/
.next/
**/*.min.js
**/*.map

# Test files (when focusing on implementation)
**/*.test.ts
**/*.test.js
**/*.spec.ts
**/__tests__/
cypress/
playwright/

# Documentation and assets
docs/
*.md
**/*.svg
**/*.png
**/*.jpg

# Configuration files
.vscode/
.idea/
*.config.js
webpack.config.js
babel.config.js

# Logs and temporary files
logs/
*.log
tmp/
temp/
.cache/</code></pre>

<h3>Step 2: Strategic Context Management</h3>
<p>Implement these strategies to maintain focused AI context:</p>

<h4>Reset Context Strategically</h4>
<p>Click the "Trash" icon in the chat header to clear the sliding window memory. Then:</p>
<ol>
    <li><strong>Be Specific:</strong> Mention exact files: "In @src/components/UserProfile.tsx, how do I..."</li>
    <li><strong>Use File References:</strong> Use @ symbol to reference specific files or folders: "@src/utils/"</li>
    <li><strong>Set Scope:</strong> Explicitly state the scope: "Focus only on the authentication logic"</li>
    <li><strong>Exclude Irrelevant:</strong> "Ignore test files and focus on the implementation"</li>
</ol>

<h4>Context Window Optimization</h4>
<pre><code>// Add to your project's .antigravity/config.json
{
  "context": {
    "max_files": 50,
    "max_lines_per_file": 1000,
    "priority_extensions": [".ts", ".tsx", ".js", ".jsx"],
    "exclude_patterns": [
      "**/node_modules/**",
      "**/dist/**",
      "**/*.test.*",
      "**/coverage/**"
    ],
    "smart_filtering": true
  }
}</code></pre>

<h3>Step 3: Workspace Organization</h3>
<p>Organize your workspace to prevent context pollution:</p>

<h4>File Structure Best Practices</h4>
<ul>
    <li><strong>Separate Concerns:</strong> Keep features, utilities, and tests in separate directories</li>
    <li><strong>Clear Naming:</strong> Use descriptive folder and file names that indicate purpose</li>
    <li><strong>Version Management:</strong> Move old versions to clearly marked archives</li>
    <li><strong>Clean Dependencies:</strong> Regularly remove unused packages and files</li>
</ul>

<h4>Advanced Context Control</h4>
<pre><code># Project-specific .antigravityignore with explanations
# Frontend assets that don't need AI analysis
public/static/**
src/assets/images/**
src/assets/fonts/**

# Third-party code
vendor/
external/
libs/third-party/

# Generated code
src/generated/**
**/*.generated.*
proto-gen/**

# Platform-specific files
.DS_Store
Thumbs.db
*.swp
*.swo

# IDE-specific files except Antigravity
.vscode/
.idea/
*.sublime-*</code></pre>

<h3>Step 4: Monitoring and Debugging</h3>
<p>Enable context debugging to understand what the AI is seeing:</p>
<pre><code>// Add to settings.json
{
  "antigravity.debug.showContextFiles": true,
  "antigravity.debug.logContextSize": true,
  "antigravity.context.verboseLogging": true
}</code></pre>

<h4>Context Health Check</h4>
<p>Regularly audit your context health:</p>
<ol>
    <li><strong>Check Context Panel:</strong> Review what files are included in each request</li>
    <li><strong>Monitor Token Usage:</strong> Track if you're hitting context limits frequently</li>
    <li><strong>Validate Responses:</strong> Ensure AI responses are relevant to your actual codebase</li>
    <li><strong>Clean Up Regularly:</strong> Remove outdated ignores and update patterns as project evolves</li>
</ol>
        

Enable Verbose Logging

Modify your Antigravity settings to capture detailed diagnostic information:

{
  "debug": {
    "logLevel": "verbose",
    "enableTimestamps": true,
    "captureStackTraces": true,
    "logToFile": true,
    "maxLogFileSize": "50MB"
  }
}

Network Connectivity Analysis

Test connectivity to Antigravity's services and dependencies:

• Check DNS resolution for api.gemini.google.com
• Verify SSL certificate validity and trust chains
• Test API endpoint availability and response times
• Analyze proxy and firewall configurations

Enterprise Deployment

For enterprise environments, consider integrating these scripts with your existing monitoring and alerting infrastructure. This enables proactive issue detection and resolution before they impact developer productivity.