Escalation Validator Skill

Validate escalation rules for correctness, completeness, and consistency.

Purpose

Ensure escalation rules are logical, cover edge cases, and have no conflicts or gaps.

Workflow

Phase 1: Load Rules

1. •

   Read Rules Document

   • •Load from docs/escalation-rules.md
   • •Identify all rule definitions
2. •

   Find Rule Engine Code

   • •Check if rules are implemented in code
   • •Locate rule evaluation logic
   • •Identify where rules are applied

Phase 2: Check Rule Logic

1. •

   Contradicting Rules

   • •Do any rules conflict?
   • •Can two rules match same input with different actions?
2. •

   Unreachable Rules

   • •Is any rule shadowed by higher priority rule?
   • •Can any rule never trigger?
3. •

   Default/Fallback Rule

   • •Is there a catch-all for unmatched cases?
   • •What happens if no rules match?
4. •

   Condition Validity

   • •Do conditions use valid field values?
   • •Are enums spelled correctly?
   • •Are ranges sensible?

Phase 3: Test Edge Cases

Create test scenarios and trace through rules:

1. •

   Boundary Conditions

   • •High priority + low churn: what happens?
   • •Sentiment exactly at threshold: which way?
   • •Multiple conditions match: correct precedence?
2. •

   Missing Fields

   • •What if optional field is null?
   • •Graceful handling or error?
3. •

   Extreme Values

   • •Sentiment = -1.0 exactly
   • •Priority = CRITICAL + high churn
   • •Issue count at exact threshold

Phase 4: Validate Actions

1. •

   Action Implementation

   • •Are all referenced actions implemented?
   • •Do action handlers exist in code?
2. •

   Required Credentials

   • •Are credentials documented in .env.example?
   • •Are missing credentials handled gracefully?
3. •

   Action Parameters

   • •Are action parameters valid?
   • •Do they match action handler signature?

Phase 5: Check Thresholds

1. •

   Frequency Thresholds

   • •Are they reasonable? (not too low/high)
   • •Based on real data?
2. •

   Sentiment Thresholds

   • •Align with scale (-1 to 1)?
   • •Make sense in context?
3. •

   No Magic Numbers

   • •Are thresholds explained/documented?
   • •Why these specific values?

Output Format

## Escalation Rules Validation

### Rules Analyzed

[List of rules by ID]

### Logical Issues

1. **[CONFLICT]** Rules R001 and R005:
   - Both match: `priority=CRITICAL AND issue_type=PRODUCT_BUG`
   - Different actions: PAGE_ONCALL vs CREATE_JIRA
   - Fix: [recommendation]

2. **[UNREACHABLE]** Rule R007:
   - Shadowed by R002 (higher priority, broader condition)
   - Fix: [recommendation]

### Edge Cases

| Scenario                             | Expected     | Actual       | Status               |
| ------------------------------------ | ------------ | ------------ | -------------------- |
| Critical bug, enterprise, high churn | Page on-call | Page on-call | ✅                   |
| Feature request, 9 mentions          | No action    | No action    | ⚠️ (threshold is 10) |

### Missing Coverage

- No rule handles: [scenario]
- Consider adding: [recommendation]

### Action Validation

| Action           | Configured         | Credentials       | Status |
| ---------------- | ------------------ | ----------------- | ------ |
| PAGE_ONCALL      | ❌ Not implemented | N/A               | ⚠️     |
| ALERT_CS_MANAGER | ✅                 | SLACK_WEBHOOK_URL | ✅     |

### Recommendations

- [Specific improvement]

Success Criteria

• • All rules checked for logical consistency
• • Edge cases tested with trace examples
• • Action implementations verified
• • Thresholds validated for reasonableness
• • Gaps in coverage identified

Constraints

• •Don't modify rules - only validate and recommend
• •Consider real-world impact - false positives vs missed escalations
• •Flag overly aggressive rules - too many alerts = alert fatigue
• •Flag overly lenient rules - critical issues slipping through

Key Files

Common Pitfalls

• •Not testing boundaries: Exact threshold values need explicit handling
• •Assuming exclusive rules: Multiple rules can match, need priority
• •Ignoring missing fields: Null/undefined fields break comparisons
• •Hardcoded values: Thresholds should be configurable

Integration with Marcus

This skill is typically invoked when:

• •Escalation rules modified
• •New escalation actions added
• •Rule engine code changes
• •Investigating escalation failures

If Blocked

If you cannot proceed:

1. •State which rules you're validating
2. •Explain what's unclear about the rule logic
3. •Provide partial validation results
4. •Request clarification on rule priority or behavior