Example Hackathon: Specifying a High-Precision AI Research Agent

[!IMPORTANT] This "Gold Standard" example focuses on using Markdown as a Logical Map. It shows how to use headings for hierarchy, lists for sequential logic, and code blocks for data contracts to generate a perfectly engineered AI system.

---

1. The Commander's Intent (High-Precision Spec)

A "10x Spec" doesn't just describe requirements; it creates a mathematical-like blueprint for the AI to follow.

Project Goal

Develop an AI Research Agent that automates the collection, synthesis, and reporting of technical documentation from multiple sources (GitHub, ArXiv, and official APIs).

Execution Constraints

1. Strict Typing: All internal data transfers must use standardized JSON schemas.
2. Modular Decomposition: The system must be separated into three distinct sub-agents: Harvester, Synthesizer, and Reporter.
3. Verification Loop: Every extracted fact must be cross-referenced across at least two sources before being included in the final report.

---

2. Structural Orchestration (Logical Hierarchy)

We use nested Markdown headers to define the system's "Decision Tree."

Phase 1: Knowledge Harvesting

• Input: Query string + list of target domains.
• Agent Action: Concurrent scraping of web documentation and PDF parsing of academic papers.
• Critical Success Criterion: 100% extraction accuracy of technical parameters.

Phase 2: Semantic Synthesis

• Action: Use a RAG (Retrieval-Augmented Generation) pipeline to identify contradictions and clusters of agreement.
• Output: A consolidated research "Graph."

---

3. The Result (High-Fidelity Artifacts)

The Architecture: Multi-Agent Interaction

We define the system flow using a Mermaid diagram to ensure the AI understands the "Hand-off" between agents.

The Data Contract: Specification by Example

Providing a "Target State" example prevents the AI from hallucinating a custom data format.

---

4. Why This is the "Best"

• Hierarchical Clarity: The use of # through ### prevents "Context Drift."
• Example-Driven: The inclusion of a JSON schema sample forces the AI to output machine-ready results.
• Zero Ambiguity: By defining the "Hand-off" points in the Mermaid diagram, we prevent the agents from getting stuck in an infinite loop.

[!TIP] Orchestrator's Lesson: When you write a spec in Markdown, you aren't just taking notes; you are writing the code of the LLM's thought process.