Autonomous Execution

Chief doesn’t just answer — it acts. Behind every Slack message, Chief runs a multi-step reasoning loop: think, pick a tool, execute, observe the result, decide what’s next.

The Loop

For a request like “Summarize last week’s sales and post it in #sales”, Chief might:

1. Plan — Decide it needs sales data → check known integrations → find Stripe.
2. Tool: stripe.list_charges — Pull last 7 days of charges.
3. Observe — 142 charges, $48,320 total.
4. Tool: bash — Run a quick aggregation script for top customers and average ticket.
5. Tool: slack.post_message — Format the summary and post to #sales.
6. Reply — Confirm in your DM with a link.

Each step is logged. You can see the full trace in Dashboard → Usage → Recent Tasks.

Self-Correction

When a tool call fails, Chief reads the error and decides what to do. Common patterns:

• Missing data → Ask the user a clarifying question rather than guess.
• API rate limit → Back off and retry with exponential delay.
• Permission denied → Check whether the credential is still valid; fall back to asking.
• Bad script → Read its own stderr, fix the bug, re-run.

You’ll see this self-correction in the thread — Chief reports what went wrong and what it’s trying instead.

Cost & Time Limits

Each task has soft and hard limits to prevent runaway loops:

If Chief hits a limit, it stops and reports the partial result. You can resume with “keep going” or end the task.

Stopping Chief

Anytime: /chief-stop or just say “Chief, stop” in the thread. Chief halts the current step, releases any held resources, and reports where it was when interrupted.