How agent escalation works
When a coding agent hits a blocker or needs human judgment, Dailybot escalates the situation into the Agent Inbox instead of guessing or stopping silently. Escalation creates a durable record: what the agent tried, what it could not verify, and what it is asking you to decide.
Common escalation triggers
- Ambiguous requirements — the agent needs you to pick between options or confirm scope.
- Missing permissions — repo, API, or chat access the automation cannot obtain.
- Policy or risk — merges, deletes, customer data, or other actions that require explicit approval.
- External dependency — a third-party system failed, timed out, or returned an error the agent cannot fix.
- Low confidence — the agent completed work but flags it for human review before delivery.
Exact triggers depend on your agent configuration and integrations; the Inbox item always states the reason in plain language.
How to respond
- Open Agent Inbox in the Dailybot web app.
- Select the escalated item and read the agent summary and any error or diff context.
- Choose the action that matches your decision: approve, request changes, reassign, or add a short instruction for the next run.
- Submit so the agent or workflow can continue; if you defer, note who owns the follow-up.
If nobody acts, the item stays open and visible to permitted roles so it does not get lost in chat noise.
Response time expectations
Targets below are guidance for teams using agents in production; your org may adopt stricter internal SLAs.
| Situation | Suggested response |
|---|---|
| Blocks a release or customer workflow | Same business day |
| Security or data-handling question | Same business day |
| General review / non-urgent | Within 2 business days |
| Informational escalation (FYI) | Acknowledge when convenient |
Dailybot does not enforce financial penalties on these windows; they exist so agents do not stall indefinitely while humans prioritize.