# Claude Code Status-Line Usage Ingestion

Pipeline tracks Claude subscription usage most accurately from Claude Code's status-line JSON. This is preferred over API rate-limit headers because Claude Code exposes provider-native subscription windows:

- `rate_limits.five_hour.used_percentage` -> Current session
- `rate_limits.five_hour.resets_at` -> Current session reset time
- `rate_limits.seven_day.used_percentage` -> All models
- `rate_limits.seven_day.resets_at` -> All models reset time

API rate-limit headers are still useful diagnostics, but they are not subscription remaining usage.

## Configure Claude Code

Add a status-line command to `~/.claude/settings.json` or `~/.config/claude/settings.json`:

```json
{
  "statusLine": {
    "type": "command",
    "command": "PIPELINE_API_URL=http://localhost:8001 PIPELINE_AUTH_TOKEN=$LOCAL_AUTH_TOKEN PIPELINE_GATEWAY_ID=<gateway-id> /home/kaspa/.openclaw/Projects/Pipeline/scripts/claude-statusline-to-pipeline.py",
    "padding": 0
  }
}
```

Replace `<gateway-id>` with the Pipeline gateway UUID shown in the dashboard/API. Use the backend URL that is reachable from the shell running Claude Code.

The collector forwards only sanitized fields: `session_id`, `model`, `workspace`, and `rate_limits`. It ignores network errors so Claude Code is not interrupted if Pipeline is down.

## Verification

After Claude Code completes at least one model response, the dashboard Runtime Usage card should show provider-native rows:

- Current session
- All models

Both should display `% used` and reset time from Claude Code. If `rate_limits` is absent, Pipeline will not invent a subscription usage percentage.