Sadd

Execution framework that dispatches fresh subagents for each task with quality gates between iterations, enabling fast parallel development while maintaining code quality.

Focused on:

• Fresh context per task - Each subagent starts clean without context pollution from previous tasks
• Quality gates - Code review between tasks catches issues early before they compound
• Parallel execution - Independent tasks run concurrently for faster completion
• Sequential execution - Dependent tasks execute in order with review checkpoints

Plugin Target

• Prevent context pollution - Fresh subagents avoid accumulated confusion from long sessions
• Catch issues early - Code review between tasks prevents bugs from compounding
• Faster iteration - Parallel execution of independent tasks saves time
• Maintain quality at scale - Quality gates ensure standards are met on every task

Overview

The SADD plugin provides skills and commands for executing work through coordinated subagents. Instead of executing all tasks in a single long session where context accumulates and quality degrades, SADD dispatches fresh subagents with quality gates.

Core capabilities:

• Sequential/Parallel Execution - Execute implementation plans task-by-task with code review gates
• Competitive Execution - Generate multiple solutions, evaluate with judges, synthesize best elements
• Work Evaluation - Assess completed work using LLM-as-Judge with structured rubrics

This approach solves the "context pollution" problem - when an agent accumulates confusion, outdated assumptions, or implementation drift over long sessions. Each fresh subagent starts clean, implements its specific scope, and reports back for quality validation.

The plugin supports multiple execution strategies based on task characteristics, all with built-in quality gates.

Quick Start

# Install the plugin
/plugin install sadd@NeoLabHQ/context-engineering-kit

# Use competitive execution for high-stakes tasks
/do-competitively "Design and implement authentication middleware with JWT support"

Usage Examples

New in v2.2 Release

Plugin was significantly improved with new agents based on LLM-as-a-Judge and LLM-as-a-Meta-Judge papers. Now it work as generalized, simplified and distiled version of Spec-Driven Development plugin. SADD plugin commands uses meta-judge agent in parallel with implementation, in order to generate in-memory specification and judge agent used to critically evaluate the implementation artifacts based on the specification.

Both judges are general purpose, so they are good as at evaluating code implementation same way as documentation, research and simple questions. As a result you should get high quality results with minimal time spend. But if you want insure aligment of code generation with your overral vision, better to use Spec-Driven Development plugin.

Commands Overview

• launch-sub-agent - This command launches a focused sub-agent to execute the provided task. Analyze the task to intelligently select the optimal model and agent configuration, then dispatch a sub-agent with Zero-shot Chain-of-Thought reasoning at the beginning and mandatory self-critique verification at the end.
• /do-and-judge - Execute a single task with implementation sub-agent, independent judge verification, and automatic retry loop until passing or max retries exceeded.
• /do-in-parallel - Execute tasks in parallel across multiple targets with intelligent model selection, independence validation, and quality-focused prompting
• /do-in-steps - Execute complex tasks through sequential sub-agent orchestration with intelligent model selection and LLM-as-a-judge verification.
• /do-competitively - Execute tasks through competitive generation, multi-judge evaluation, and evidence-based synthesis to produce superior results.
• /tree-of-thoughts - Execute complex reasoning tasks through systematic exploration of solution space, pruning unpromising branches, expanding viable approaches, and synthesizing the best solution.