Monitor & Start
Metateam is designed around automatic hooks. monitor and start are support tools for clients or environments where hooks are unavailable or when you need to debug context injection.
Monitor
$ metateam monitor [AGENTS]
Runs a background process that watches for AI agent sessions and uploads their transcripts when they finish.
| Argument | Default | Purpose |
|---|---|---|
codex |
yes | Monitor Codex sessions |
gemini |
Monitor Gemini sessions | |
claude |
Monitor Claude sessions (usually handled by hooks instead) |
Examples:
$ metateam monitor # Monitor Codex (default)
$ metateam monitor codex gemini # Monitor both Codex and Gemini
How It Works
The monitor:
- Polls for running agent processes at regular intervals
- When an agent process is detected, opens a time window
- When the process exits, scans the agent's transcript directory for files modified during that window
- Uploads any new transcripts in the background
This is best-effort. It depends on:
- Being able to detect the agent process (process name/PID matching)
- Finding transcript files in the expected location
- The transcript being complete when the process exits
For interactive setup and hook configuration, use metateam setup and metateam hook.
Start
$ metateam start [PROMPT]
Manually triggers context injection for sessions where hooks are not available. Outputs the same context that SessionStart hooks produce. This is rarely needed -- hooks handle context injection automatically for all supported clients.
An agent (or user in rare cases) can invoke this to retrieve relevant past sessions and KB entries as injectable context.
| Option | Purpose |
|---|---|
--as <IDENTITY> |
Override identity as station:crew:persona (for debugging recall and context selection) |
Example workflow:
$ metateam start "fix the auth module" > context.txt
$ metateam start --as bronto:titan:data "review this architecture"
If no prompt is provided, it uses an empty string and returns context based on the working directory and git metadata alone.