CortexPrism Chain-of-Thought Agent Strategy

Pluggable reasoning strategies that Cortex agents can dynamically select per-task for smarter, more structured problem-solving. Includes Chain-of-Thought, Tree-of-Thoughts, ReAct, and Plan-and-Execute strategies.

Installation

cortex plugin install cortex-plugin-chain-of-thought

Or install from local development:

git clone https://github.com/CortexPrism/cortex-plugin-chain-of-thought.git
cd cortex-plugin-chain-of-thought
cortex plugin install .

Reasoning Strategies

Strategy	Best For	Description
Chain-of-Thought (CoT)	Math, Logic, Debugging	Linear step-by-step reasoning through a problem
Tree-of-Thoughts (ToT)	Planning, Creative, Decision	Branching exploration of multiple reasoning paths with pruning
ReAct	Coding, Debugging, Decision	Interleaved reasoning and action with observation cycles
Plan-and-Execute	Planning, Coding, Analysis	Two-phase: create detailed plan, then execute in dependency order

Configuration

Setting	Type	Default	Description
defaultStrategy	string	auto	Default strategy when none specified (cot, tot, react, plan_execute, auto)
maxTreeDepth	number	5	Maximum depth for Tree-of-Thoughts branching
treeBreadth	number	3	Number of branches per node in Tree-of-Thoughts
reactMaxIterations	number	15	Maximum thought-action-observation cycles for ReAct

Tools

reason

Execute a reasoning strategy on a given problem.

{
  "problem": "How do I optimize this database query that's taking 5 seconds?",
  "strategy": "react",
  "context": "SELECT * FROM users JOIN orders ON ...",
  "max_steps": 10
}

list_strategies

List all available reasoning strategies with descriptions and recommendations.

{}

select_strategy

Auto-select the best strategy for a task.

{
  "task_description": "Debug a failing CI pipeline",
  "task_type": "debugging"
}

evaluate_reasoning

Evaluate the quality of a reasoning trace for logical gaps and coherence.

{
  "reasoning_trace": "Step 1: ... Step 2: ...",
  "criteria": "logic,completeness,clarity"
}

Usage Example

> I need to solve a complex architecture decision

1. select_strategy → { task_type: "decision" }
→ Selected: Tree-of-Thoughts

2. reason → { strategy: "tot", problem: "Monolith vs Microservices for our app..." }
→ Explores 3 branches, prunes 2, identifies best path with 0.82 confidence

How It Works

This plugin operates as a middleware layer in the Cortex agent loop:

1. Pre-middleware: Before each agent turn, the plugin can inject the selected strategy's system prompt
2. Tool execution: The reason tool simulates running the strategy (in production, this delegates to the LLM with strategy-specific prompting)
3. Post-middleware: After each agent turn, the plugin can evaluate reasoning quality and suggest strategy adjustments

Development

deno task test
deno fmt && deno lint

# Test reasoning
cortex plugin call cortex-plugin-chain-of-thought reason '{"problem":"What is 2+2?","strategy":"cot"}'

License

MIT