ATIF Preview

See what your AI agent actually did. Open any ATIF trajectory file and get an interactive, step-by-step view of every message, tool call, reasoning trace, and token cost — right inside VS Code.

Why

Raw trajectory JSON is painful to read. Scrolling through hundreds of lines to find a failed tool call or understand an agent's reasoning chain kills your flow. This extension turns that JSON into a navigable timeline so you can diagnose agent behavior in seconds instead of minutes.

Features

• One-click preview — click the preview icon in the editor title bar on any JSON file
• Step-by-step timeline — every user message, agent response, and system event as a collapsible card
• Tool call inspection — expand any tool call to see arguments, results, and matched observations
• Reasoning traces — agent thinking/chain-of-thought displayed inline with visual distinction
• Token & cost metrics — per-step and aggregate prompt tokens, completion tokens, cached tokens, and USD cost
• Multi-trajectory support — files with multiple trajectories render as tabs (parent + subagent)
• Embedded subagent trajectories — v1.7 subagent_trajectories arrays are flattened into the tab view
• Subagent navigation — click through to linked subagent trajectories via trajectory_id (embedded) or external file references
• LLM call count — badges for deterministic (no-LLM) steps and multi-LLM aggregated steps
• Copied context — steps marked is_copied_context are visually dimmed with a badge
• Filter & search — filter steps by tool name or search across all message content
• Skill usage detection — spot when an agent reads files from a known Agent Skills location (.github/skills/, .claude/skills/, .agents/skills/, .copilot/skills/), with per-skill filter pills and step indicators
• Live reload — preview updates automatically as you edit the source file
• Native VS Code look — uses your theme colors, codicons, and standard UI patterns

Getting Started

1. Install the extension
2. Open an ATIF trajectory JSON file
3. Click the preview icon in the editor title bar (or run ATIF: Preview Trajectory from the Command Palette)

The preview opens in a side panel so you can read the trajectory alongside the raw JSON.

What is ATIF?

Agent Trajectory Interchange Format (ATIF) is an open schema for recording what AI agents do — the messages, tool calls, observations, metrics, and reasoning that make up a session. It gives you a standard way to capture, share, and analyze agent runs across different frameworks and models.