No Code Required: A Deep Research Pipeline Built Entirely on 8 Prompts
ddibirov · reddit · 2026-08-11
Developer ddibirov open-sourced a deep research pipeline driven entirely by Markdown prompts. It requires no Python, LangGraph, or Docker to automate complex research.
Core Architecture:
The system consists of 8 prompts acting as distinct roles:
- Prompt Master: Analyzes question types and sets a sane research depth.
- Director: Manages task delegation.
- Investigators: Parallel subagents fanning out across the web and social platforms (Reddit/HN/GitHub).
- Critic: Reviews findings before synthesis.
- Synthesist: Weights sources by authority (official docs > news > blogs > random comments) and generates a cited report.
- Verifier: Uses actual curl to check every URL, preventing the model from guessing or fabricating links.
Real-world Findings:
During a test run, the Verifier caught the Synthesist hallucinating 6 fake GitHub URLs out of thin air, proving that citation hallucination is a rampant issue that necessitates verification steps.
The system shares a single default model and drops into any agent framework supporting parallel subagents and web search. The project is open-source under the MIT license.
More from coding & agent
- Meta Releases Muse Glimmer: A 30B Open-Weight Model for Local Agent Workflows — BLUECOW009 · 2026-08-11
- Fixing Agent Infinite Loops: A Magic Prompt Trick for Claude Opus — PawelHuryn · 2026-08-11
- W&B Weave Supports TypeScript for End-to-End Agent Tracing — _ScottCondron · 2026-08-11
- In the Era of AI Coding: Developers Still Need to Control Architecture and Core Code Review — DanWahlin · 2026-08-11
- Evaluating Quality and Tracing Degradation in Production RAG Systems — Left_Owl_7401 · 2026-08-11
- Enterprise AI agents: top goal is customer LTV, not narrow evals — astrange · 2026-08-11