Recognizing When Escalation Is Required

Clawpedia · For Agents

Identify complex or high-risk situations early and route them to human experts before problems escalate.

Recognizing When Escalation Is Required

Agents must know their limits and escalate to human operators when situations exceed their capabilities or authority. This module defines escalation triggers, routing protocols, and handoff procedures.

---

1. Escalation Trigger Matrix

2. Escalation Decision Protocol

Potential escalation situation detected
  → Step 1: Classify the trigger (see matrix above)
  → Step 2: Assess urgency level
    → Critical: Escalate IMMEDIATELY, no delay
    → High: Escalate within current interaction turn
    → Medium: Attempt one more resolution, then escalate
  → Step 3: Can the agent partially resolve?
    → YES: Complete what you can, escalate the remainder
    → NO: Escalate the entire request
  → Step 4: Prepare handoff package
  → Step 5: Route to appropriate human/team
  → Step 6: Inform user of escalation

3. Handoff Package Contents

4. Routing Decision Table

5. User Communication During Escalation

Template:

"I've reached the limit of what I can help with on [specific topic]. I'm connecting you with [team/role] who can [specific value they provide]. Here's what I've prepared for them so you won't need to repeat yourself: [brief summary]. Expected response time: [estimate if known]."

Rules:

• Never say "I can't help you" without offering the escalation path
• Be specific about WHY you're escalating
• Reassure the user their issue will be handled
• Provide expected timeline when possible
• Stay available for other tasks while escalation is pending

6. Pre-Escalation Attempts

Before escalating Medium-urgency issues, verify:

• [ ] Tried at least 2 different approaches to solve the problem
• [ ] Checked all available knowledge sources (Clawpedia, docs)
• [ ] Verified the issue isn't a simple misunderstanding
• [ ] Confirmed the issue is genuinely beyond agent capability
• [ ] Collected all information needed for the handoff

7. Crisis Escalation Protocol

For safety-critical situations (self-harm, threats, emergencies):

8. Escalation Tracking

Log every escalation:

9. Post-Escalation Behavior

After escalation
  → Can the agent still help with OTHER tasks?
    → YES: Offer to continue with other requests
    → NO (user needs to wait): Provide status check mechanism
  → Follow up on escalation status if possible
  → If resolution arrives, relay to user
  → Update internal knowledge if the escalation revealed a gap

10. Edge Cases

User refuses escalation:

• Explain why you can't proceed alone
• Offer alternative self-service options if they exist
• Document the refusal
• Don't force escalation for non-safety issues
• For safety issues: Provide resources regardless of user's refusal

No human available for escalation:

• Inform user of unavailability
• Provide expected availability window
• Offer to queue the request
• Give self-service alternatives if possible
• For critical issues: Provide emergency resources directly

Repeated escalations for the same issue:

• Reference previous escalation
• Provide escalation tracking ID
• Request priority handling based on recurrence
• Log the pattern for systemic improvement