/review-plan

Agent Prompt Templates

Full prompt templates for the reviewers and nested workflows selected in Step 3. Apply that step's tool restrictions and coordination limits.

External Opinions Agent

Invoke the second-opinion skill with this prompt:

Review this implementation plan:

{plan_summary}

Key questions:
1. Is this the right approach?
2. What are we missing?
3. Any red flags?

Alternatives Agent

Prompt:
---
Given this problem and proposed solution:

**Problem:** {problem_statement}
**Proposed Solution:** {solution_summary}

Propose {2-4} alternative approaches. For each:

## Alternative {N}: {Name}

**Approach:** {brief description}

**Pros:**
- {advantage 1}
- {advantage 2}

**Cons:**
- {disadvantage 1}
- {disadvantage 2}

**When to prefer:** {scenarios where this is better}

Focus on meaningfully different approaches, not minor variations.
---

Robustness Agent

Prompt:
---
Review this plan for robustness issues:

{full_plan}

Check for these anti-patterns (see references/robustness-patterns.md for examples):

## Timing-Based "Solutions" (RED FLAGS)
- Timeouts to "fix" race conditions (use proper synchronization)
- Sleep/delay to wait for async operations (use await/callbacks/events)
- Polling when events/subscriptions are available
- Arbitrary delays hoping state settles

## Error Handling Issues
- Swallowing errors silently
- Catch-all without specific handling
- Missing rollback/cleanup on failure
- No retry strategy for transient failures

## State Management Issues
- Global mutable state without synchronization
- Optimistic updates without conflict resolution
- Cache invalidation assumptions
- Stale closure captures

## Concurrency Issues
- Shared state without locks/atomics
- Missing transaction boundaries
- Fire-and-forget async without error handling
- Assumption of execution order

## Scalability Issues
- O(n^2) or worse algorithms on unbounded data
- Loading all data into memory
- No pagination/streaming for large datasets
- Blocking operations in event loops

For each issue found:
1. Quote the problematic part of the plan
2. Explain why it's fragile
3. Suggest a robust alternative
---

Adversarial Agent

Prompt:
---
Be maximally critical of this plan. Your job is to find flaws.

{full_plan}

Attack from every angle:

**Correctness:** Will this actually solve the problem? Edge cases?

**Completeness:** What's missing? What will break?

**Complexity:** Is this overengineered? Underengineered?

**Maintainability:** Will future developers understand this? Will it rot?

**Testing:** How will we know it works? What's hard to test?

**Deployment:** What could go wrong in production?

**Dependencies:** Are we relying on something fragile?

**Assumptions:** What are we assuming that might not be true?

Be harsh. Better to find problems now than after implementation.
Do not soften criticism. If something is bad, say it's bad.
---

Research Agent

Invoke the research-tech skill with relevant topic:

{library/technology mentioned} {core problem} best practices

Focus on:
- Known issues with proposed approach
- Best practices we might be missing
- Recent changes that affect the plan

Spawning All Agents (Example)

Schedule the three reviewers and two skill workflows in parallel as runtime capacity allows:

1. second-opinion skill with plan summary
2. restricted worker for alternatives
3. restricted worker for robustness
4. restricted worker for adversarial
5. research-tech skill with relevant topic

Wait for ALL to complete before proceeding.

Verify before Step 4:

• External Opinions agent spawned (second-opinion)
• Alternatives agent spawned
• Robustness agent spawned
• Adversarial agent spawned
• Research agent spawned (research-tech)