Event Reviewer

Quality-check outputs from multiple event skills for consistency, completeness, and feasibility. Catches the gaps that individual skills cannot see because they operate independently.

When This Skill Activates

• User has outputs from 2+ event skills and wants a consistency check
• User asks to "review my event plan" or "check everything fits together"
• Before presenting a complete event plan to stakeholders
• After a major change to one skill's output that might affect others
• User asks if anything is missing or inconsistent

This skill does NOT: Generate event deliverables (use specialist skills). Does not replace individual skill verification gates (Step 3.5). Reviews cross-skill consistency that individual skills cannot check alone.

Step 1: Gather Outputs

Collect all available outputs from event skills. At minimum, need outputs from 2+ skills to provide value.

Step 2: Run Consistency Checks

Machine-readable thresholds and weights for automated evaluation are available in tests/evaluation-rubrics.json.

Check 1: Numbers Alignment

Check 2: Timeline Alignment

Check 3: Budget Alignment

Check 4: Capacity Alignment

Check 5: Data Protection Compliance

Validates regional data protection rules:

• EU/EEA events: GDPR — double opt-in for DE/AT, standard consent elsewhere
• Singapore/APAC: PDPA — opt-in consent, broadly mirrors GDPR
• US events: State laws where applicable (CCPA for California residents)
• Default: Apply GDPR-equivalent consent as baseline for any region

Format-Aware Check Adjustment:

• Virtual events: Skip venue capacity, catering count, load-in schedule, and booth allocation checks
• Hybrid events: Apply venue/catering checks to in-person count only (use attendee_count_in_person if available, not total attendee_count)
• In-person events: All checks apply

Check 6: Content Completeness

Step 3: Generate Review Report

Report output format:

## Review Summary
- **Status**: PASS / WARN / FAIL
- **Checks run**: X/6
- **Critical issues**: [count]
- **Warnings**: [count]

## Detailed Results
| Check | Rule | Expected | Actual | Status |
|---|---|---|---|---|
| Numbers | Attendee count | 500 across all skills | 500 (planner), 480 (budget) | WARN |

## Recommended Actions
1. [CRITICAL] Align attendee count — update event-budget to 500
2. [WARN] Add contingency to venue budget — use the event-budget skill

## Prescriptive Remediation
For each CRITICAL/WARN finding, provide:
| # | Finding | Fix Command | Skill | Input Change | Expected Result |
|---|---|---|---|---|---|
| 1 | Attendee mismatch (500 vs 480) | Re-run event-budget with attendee_count=500 | event-budget | attendee_count: 500 | Budget recalculated for 500 |
| 2 | No contingency | Add 10% contingency line | event-budget | Add contingency: 10% | Budget + 10% contingency |

For each finding: cite the specific values from each skill, state the impact, and name the skill responsible for the fix. The prescriptive remediation table must specify the exact input change needed — not just "fix it" but "change X to Y in skill Z."

Step 3.5: Verification Gate

1. All provided outputs reviewed: No skill output was skipped
2. Issues are specific: Each finding cites exact numbers/dates, not vague concerns
3. Recommendations are actionable: Each issue has a clear fix with responsible skill identified
4. Priority is clear: Critical vs. warning vs. gap distinction maintained

Step 4: Cross-Skill References

• Fix budget issues: "For budget adjustments, use the event-budget skill"
• Fix timeline issues: "For timeline changes, use the event-planner skill"
• Fix content gaps: "For missing content, use the event-content skill"
• Orchestrate fixes: "For coordinating changes across multiple skills, use the event-orchestrator skill"

Output Format

• Review summary: status (PASS / WARN / FAIL), check counts, critical and warning counts
• Detailed results table: Check / Rule / Expected / Actual / Status — cite exact values from each skill
• Recommended actions: numbered list, prefixed with [CRITICAL] / [WARN] / [GAP], each with a fix command and responsible skill
• Prescriptive remediation table: # / Finding / Fix Command / Skill / Input Change / Expected Result
• No new deliverables generated — reviewer only validates existing outputs

Examples

Example 1: Budget-timeline mismatch

User provides: Event brief (500 attendees), budget (150K), speaker plan (30 speakers with fees) Finding: Speaker budget allocates 33% of total for 30 speakers (avg 1,667 per speaker), but 5 keynotes at 3-7% of total budget each consume 17-33% alone, leaving 0-17% for 25 remaining speakers. Recommendation: Adjust speaker budget to ~47% of total or reduce keynote count to 3.

Example 2: Registration-marketing gap

User provides: Registration plan (early-bird ends T-6 weeks), marketing plan (promotion starts T-8 weeks) Finding: Only 2 weeks between promotion start and early-bird deadline. Benchmark: 4-6 weeks needed for early-bird conversion. Recommendation: Either extend early-bird to T-4 weeks or start promotion at T-10 weeks.

Example 3: GDPR inconsistency

User provides: Sponsor package (Platinum gets full attendee list), registration form (no sponsor data-sharing consent checkbox) Finding: Platinum tier promises attendee data but registration form has no consent mechanism. This is a GDPR violation for EU events. Recommendation: Add explicit sponsor data-sharing consent to registration form with named sponsors.

Edge Cases

• Only 2 skills provided: Run available checks, note which cross-checks cannot be performed, recommend completing other skills first.
• Outputs from different time periods: Flag if outputs may be based on different assumptions (e.g., budget from January, speaker plan from March with different attendee target).
• User disagrees with finding: Explain the reasoning, cite the benchmark, but respect user's final decision. Document the accepted risk.
• No issues found: Confirm consistency and list which checks passed. Recommend areas to monitor as plans evolve.
• Major inconsistency cascade: If one change affects 3+ other skills, recommend using event-orchestrator to manage the update sequence.