Trajectory System

The Trajectory System is OmniIDE's action logging infrastructure. Every step the Omni-Agent takes — every file write, command execution, and code generation — is recorded in a structured trajectory that can be reviewed, debugged, and replayed.

What Gets Recorded

• Agent instructions (user prompts)
• Generated execution plans
• File creation and modification events
• Terminal commands and their outputs
• Error detection and healing attempts
• Final success or failure state

Trajectory Structure

{
  "sessionId": "abc-123",
  "timestamp": "2026-06-01T10:00:00Z",
  "steps": [
    { "type": "plan", "content": "Create Express server" },
    { "type": "file_write", "path": "server.js", "content": "..." },
    { "type": "terminal", "command": "npm install express", "exitCode": 0 },
    { "type": "terminal", "command": "node server.js", "exitCode": 0 },
    { "type": "success", "message": "Server running on port 3000" }
  ]
}

Use Cases

• Debugging — Review exactly what the agent did when something went wrong
• Learning — Understand the agent's decision-making process
• Auditing — Track all autonomous operations for compliance

Related

• Memory Architecture — How the agent retains context
• Verification Engine — Error detection system