S02 — GitHub System Prompts Search Log¶
Tool and Description¶
| Field | Value |
|---|---|
| Tool | WebSearch |
| Query | GitHub published AI system prompt structured analytical framework research agent competing hypotheses evidence scoring |
| Date | 2026-03-31 |
| Results returned | 10 |
| Selected | 4 |
| Rejected | 6 |
Additional searches: awesome-system-prompts research methodology, OpenAI deep research system prompt leaked, Perplexity system prompt leaked
Summary¶
Searched GitHub repositories and prompt collections for published system prompts implementing analytical frameworks. Found extensive collections of leaked/published commercial prompts (Perplexity, OpenAI, Claude, etc.) but none implementing comprehensive research methodology. Found partial implementations of SATs via LLMs.
Selected Results¶
| Result | Title | Rationale |
|---|---|---|
| S02-R01 | sroberts LLM-SATs-FTW | Implements 3 structured analytic techniques via LLM |
| S02-R02 | Perplexity AI system prompt (leaked) | Commercial research tool — no analytical framework |
| S02-R03 | OpenAI Deep Research system prompt (leaked) | Commercial deep research tool — no analytical framework |
| S02-R04 | awesome-ai-system-prompts collection | Comprehensive prompt collection — no research methodology prompts |
Rejected Results¶
| Result | Title | Rationale for Rejection |
|---|---|---|
| — | x1xhlol/system-prompts-and-models-of-ai-tools | Collection of coding/chat prompts, no research methodology |
| — | HKUDS/AI-Researcher | Covered under S03 preprint search |
| — | assafelovic/gpt-researcher | Tool, not published prompt with analytical framework |
| — | tmgthb/Autonomous-Agents | Paper collection, not prompts |
| — | promptslab/Awesome-Prompt-Engineering | General prompt engineering, no research methodology |
| — | elder-plinius/CL4R1T4S | Leaked prompts collection — examined, no research frameworks found |
Notes¶
The gap between commercial system prompts (which contain no analytical methodology) and academic research (which discusses methodology but does not publish complete prompts) is the central finding. Commercial prompts delegate analytical rigor to model training and retrieval architecture, not to prompt design.