By name: workflow-review description: "Run parallel multi-agent review on completed knowledge work. Use after the work phase to evaluate output quality from multiple specialized perspectives before finalizing. Triggers on requests like 'review this', 'check this before I send it', 'stress-test this recommendation', or when quality assurance is needed on a deliverable."
Review: Multi-Agent Knowledge Work Review
You are orchestrating the Review phase of the Compound Knowledge Work loop. Your job is to evaluate completed work from multiple specialized perspectives and synthesize findings into actionable improvements.
Different perspectives catch different issues. A single reviewer misses things that parallel review catches.
Process
Phase 1: Determine Review Scope
- Read the completed work thoroughly.
- Identify the work type to select appropriate reviewers:
| Work Type | Standard Reviewers | Conditional Reviewers |
|---|---|---|
| Communication | clarity-reviewer, tone-calibrator, completeness-auditor | sensitivity-scanner (if sensitive topic), strategic-alignment-checker (if strategic) |
| Decision | completeness-auditor, risk-assessor, devil's-advocate, actionability-validator | strategic-alignment-checker (if organizational), sensitivity-scanner (if political) |
| Analysis | completeness-auditor, clarity-reviewer, actionability-validator | risk-assessor (if recommendations included), devil's-advocate (if controversial) |
| Meeting prep | completeness-auditor, actionability-validator, clarity-reviewer | stakeholder-mapper (if complex dynamics), sensitivity-scanner (if contentious topics) |
| Coaching notes | clarity-reviewer, actionability-validator | tone-calibrator (if shared with coachee) |
- Check if the plan specified reviewers. If so, use those as the baseline.
Phase 2: Parallel Agent Review
Launch all selected review agents in parallel using the Task tool.
Each agent reviews independently and returns findings in their specialized format.
Agent Instructions:
- Each agent receives the complete work output
- Each agent applies its specialized lens
- Each agent returns findings with severity levels
Phase 3: Synthesize Findings
After all agents return, synthesize their findings:
Output Format:
## Review Summary
### Overall Assessment
**Quality Score**: [1-10]
[2-3 sentence assessment of overall quality and readiness]
### Critical Issues (Must Fix)
Issues that would cause real problems if not addressed.
#### Issue 1: [Title]
- **Found by**: [Agent name]
- **Location**: [Where in the document]
- **Problem**: [What's wrong]
- **Impact**: [What happens if not fixed]
- **Fix**: [Specific recommendation]
### Important Issues (Should Fix)
Issues that meaningfully improve quality.
#### Issue 1: [Title]
- **Found by**: [Agent name]
- **Location**: [Where in the document]
- **Problem**: [What's wrong]
- **Fix**: [Specific recommendation]
### Minor Issues (Nice to Fix)
Polish and refinement opportunities.
- [Issue]: [Quick fix] _(found by [agent])_
- [Issue]: [Quick fix] _(found by [agent])_
### What's Working Well
- [Strength 1] _(noted by [agent])_
- [Strength 2] _(noted by [agent])_
### Conflicting Feedback
[If agents disagree, note the conflict and recommend resolution]
### Recommended Action
- [ ] Fix critical issues (estimated: [time])
- [ ] Fix important issues (estimated: [time])
- [ ] Fix minor issues (estimated: [time])
- [ ] Ready to finalize after fixes
Phase 4: Apply Fixes
If the user approves:
- Fix critical issues first.
- Fix important issues.
- Fix minor issues if time permits.
- Present the revised version with a summary of changes made.
If issues conflict, present the tradeoff to the user and let them decide.
Review Principles
Severity Calibration
- Critical: Would cause real damage - wrong tone in sensitive email, missing key stakeholder concern, factual error in analysis, flawed logic in decision
- Important: Meaningful quality improvement - unclear structure, missing context, weak evidence, incomplete consideration
- Minor: Polish - word choice, formatting, small redundancies
Independence Matters
- Each agent reviews independently
- Don't let one agent's findings influence another
- Conflicts between agents are valuable signal
Review Should Be Proportional
- Quick internal note: clarity-reviewer only
- Important email: clarity + tone + completeness
- Board presentation: all standard + conditional reviewers
- Strategic decision: all reviewers including devil's-advocate
The Devil's Advocate is Special
- Always include for decisions and recommendations
- Its job is to find the strongest counterargument
- Don't dismiss its findings just because you disagree
- If the devil's advocate can't find a good counterargument, that's strong validation
Next Step
When the work is finalized, capture what you learned: /coworkpowers:workflow-compound
Anti-Patterns to Avoid
- Running review on low-stakes work (wastes time)
- Ignoring conflicting feedback instead of resolving it
- Treating all findings as equal severity
- Over-revising based on minor feedback
- Not running devil's advocate on important decisions
- Applying fixes without user awareness of what changed