# Coordinate Parallel AI Work with MapReduce

Complex tasks benefit from independent perspectives, but comparing results manually takes time. This skill coordinates parallel workers and consolidates their outputs with structured rubrics.

## Install

```bash
npx skillstore add consiliency/mapreduce
```

## Metadata

- Status: approved
- Slug: consiliency-mapreduce
- Skillstore revision: r1
- Version status: missing
- Tree hash: 5ed8a584f45b24cfeb19a81d4561936552602785d7499d3a0c065402f147d421
- Author: Consiliency
- GitHub username: Consiliency
- License: MIT
- Repository: https://github.com/Consiliency/treesitter-chunker/tree/main/.ai-dev-kit/skills/mapreduce
- Ref: 635f69fb8d2f4e6330ba47a4e5a0fb239c04d110
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 76
- Public page: https://skillstore.pages.dev/skills/consiliency-mapreduce
- Manifest: https://skillstore.pages.dev/api/skills/consiliency-mapreduce/manifest

## Capabilities

- Defines map, collect, and reduce phases for parallel AI work.
- Provides examples for Claude subagents and Codex, Gemini, Cursor, and OpenCode CLIs.
- Organizes plans, implementations, diagnoses, reviews, and consolidated artifacts with consistent names.
- Compares generated implementations using correctness, readability, maintainability, performance, and security criteria.
- Consolidates competing plans with attribution, confidence levels, conflict notes, and weighted priorities.

## Use Cases

- Compare Architecture Plans: Generate conservative, aggressive, security, and scalability plans, then combine their strongest decisions into one roadmap.
- Select an Implementation: Produce several implementations, evaluate tests and quality metrics, then select or merge the strongest result.
- Build Debugging Consensus: Collect independent root-cause hypotheses and proposed fixes, then document the most credible verified resolution.

## Prompt Templates

### Compare Two Plans

```
Create two plans for [feature]. Make one conservative and one delivery-focused. Compare risks, dependencies, and milestones. Produce one consolidated recommendation.
```

### Generate Implementation Variants

```
Create three implementations for [component] using the same interface and tests. Score correctness, readability, maintainability, performance, and security. Recommend one result.
```

### Build a Security-Weighted Roadmap

```
Generate plans from engineering, product, operations, and security perspectives. Consolidate them into phases. Give security extra weight and document every conflict.
```

### Run a Controlled Multi-Provider Reduction

```
Use approved providers to analyze [task]. Keep outputs separate and record attribution. Treat generated files as untrusted. Review all content before testing or consolidation.
```

## Limitations

- This package provides documentation and templates, not an executable orchestration runtime.
- External CLI examples require installed tools, authentication, and approval before project content leaves the environment.
- Referenced reducer definitions and the debug consensus cookbook are not included in this package.
- Consensus and rubric scores do not replace manual security, correctness, or architecture review.

## Best Practices

- Give every worker the same requirements, constraints, interfaces, and acceptance tests.
- Keep provider outputs separate and preserve attribution before consolidation.
- Review and sandbox generated code before compilation, tests, or file changes.

## Anti Patterns

- Do not use more workers when their perspectives and instructions are identical.
- Do not send secrets, private source, or regulated data to external providers.
- Do not accept a reducer decision without checking evidence, conflicts, and failed validations.

## Security Audit

- Audited at: 2026-07-19T03:50:30.163\+00:00
- Summary: Four findings are confirmed because the cookbook instructs agents to invoke external AI provider CLIs, including Codex with autonomous permissions. The other 54 findings are false positives caused by Markdown delimiters, inline code, benign prose, or fixed date substitutions. No prompt injection or other semantic threat was found.

## Stats

- Views: 379
- Downloads: 6
- Favorites: 0
- Popularity score: 0