Research Firm Revealed Coding Agent System Prompts
Novel Cognition published 23 captured system prompts used by nine prominent AI coding agents.
Updated on Sept. 20, 2026 in Artificial Intelligence
Live Poll
Do you trust AI coding tools that share verbatim instruction prompts with their competitors?
Novel Cognition has released a public repository featuring 23 system prompts captured from nine different AI coding agents. The analysis highlights significant overlaps in how these agents are instructed to perform tasks.
Why it matters
By exposing the hidden scaffolding instructions used by AI models, this research reveals the extent to which different coding agents rely on identical prompt templates. This transparency allows the public to better understand the underlying logic governing various automated coding tools.
The repository includes prompts ranging from 1,954 to 50,399 characters. Notably, four agents share a 147-character verbatim sentence, while OpenCode utilizes a 114-line template across five models.
The players
Novel Cognition
A research entity that conducts technical analysis on the internal operations and instructions of artificial intelligence systems.
OpenCode
An AI coding agent platform that was found to use a standardized 114-line prompt template across five different models.
The details
Novel Cognition captured these prompts at the wire during transmission between the harness and the AI models. Line-by-line diffs revealed that Kilo Code and OpenCode share 103 lines of instruction, while MiMoCode and MiMoCode Pro utilize byte-identical system prompts.
Timeline
Novel Cognition published the coding agent prompt corpus in September 2026.
The Tech Race
This release challenges the industry tendency to maintain proprietary secrecy over AI scaffolding instructions. By mapping these templates, the research exposes the lack of architectural diversity currently shaping the competitive landscape of coding agents.
Developers and users can now compare how different AI coding agents are instructed to behave, potentially influencing tool selection based on prompt transparency. This data helps users understand if their preferred coding assistant relies on generic templates or unique instructions.
The takeaway
The ubiquity of identical prompt structures suggests that many coding agents are built on highly similar operational foundations. Understanding these shared instructions can help users evaluate the distinct value and reliability of different AI tools.
Further reading
Learn more about the evolving landscape of AI development in our Artificial Intelligence section.
More information
View the full coding agent prompt corpus analysis to explore the repository data.
Source note: This article includes information reported by The Cherry Creek News.
Live Poll
Do you trust AI coding tools that share verbatim instruction prompts with their competitors?
