Cognitrace
Cognitrace automatically logs all interactions with AI coding assistants — prompts and responses — into a structured JSON file inside your project. No configuration required.
Supported assistants
- Claude Code (VS Code extension and CLI)
- GitHub Copilot Chat
- OpenAI Codex CLI
How it works
When activated, Cognitrace watches the session files that each AI assistant writes locally and captures every user message and assistant response. Logs are written to a .ai_log/ folder at the root of your workspace.
Each log file is named prompt_log_YYYY-MM-DD_<git-username>.json and contains an array of entries:
[
{
"timestamp": "2025-04-12T10:32:00.000Z",
"source": "claude",
"git_branch": "main",
"git_user_name": "Jane Doe",
"call_context": {
"cwd": "/Users/jane/projects/my-app"
},
"role": "user",
"content": "refactor the auth module to use JWT"
},
{
"timestamp": "2025-04-12T10:32:05.000Z",
"source": "claude",
"git_branch": "main",
"git_user_name": "Jane Doe",
"call_context": {
"cwd": "/Users/jane/projects/my-app"
},
"role": "assistant",
"content": "Here's the refactored auth module using JWT..."
}
]
| Field |
Description |
timestamp |
ISO 8601 timestamp of when the message was captured |
source |
Which assistant produced the message (claude, github_copilot, codex) |
git_branch |
Active git branch at the time of the interaction |
git_user_name |
Git user name from the workspace config |
call_context.cwd |
Absolute path of the workspace |
role |
user for prompts, assistant for AI responses |
content |
The message text |
Output channel
Open View > Output and select Cognitrace from the dropdown to see real-time diagnostic logs — which directories are being watched, which files are detected, and which messages are captured.
Notes
- The
.ai_log/ folder is created automatically. Add it to .gitignore if you don't want logs committed.
- Logs are append-only and split by day and git user, so parallel sessions don't overwrite each other.
- No data is sent anywhere. Everything stays on disk, locally.
