Jean2 AI Coding Agent Launches With Fully Transparent, User-Controlled System Prompts

·3 views

Most AI coding agents such as Claude Code, Cursor, and Copilot ship with hidden system prompts that users cannot view or modify, making unexpected behavior difficult to debug. These baked-in instructions can silently override user preferences, such as package manager choices, without any visible explanation. Jean2 is a new AI coding agent built with no default system prompt or tools, giving users complete control over how the agent is configured. Its system prompt is assembled from user-defined files including a preconfig prompt, a project-level AGENTS.md, workspace memory, and loadable skill playbooks. The approach aims to eliminate the conflict between hidden product instructions and user intent, making agent behavior fully traceable and auditable.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

OpenAI Creates Public Fork of Git Version Control System on GitHub

OpenAI has published a fork of the Git version control system on its official GitHub profile. The repository is publicly accessible under the OpenAI GitHub organization. The move was noted by the developer community on Hacker News, though details about the fork's specific purpose remain limited. It is currently unclear whether the fork introduces custom modifications for internal tooling or AI-related development workflows.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Software Engineer Begins 32-Week Journey From AI APIs to Local LLM Systems on NVIDIA DGX Spark

A software engineer and technical program manager has launched a 32-week self-directed learning series aimed at transitioning from consuming AI via APIs to building and optimizing local large language models. In the first week, the author focused on understanding the hardware environment of an NVIDIA DGX Spark machine, using command-line tools like nvidia-smi and lscpu to verify system specs rather than relying on assumptions. The DGX Spark runs on an ARM64 architecture with a 20-core CPU, while the author's control machine is an Intel x86_64 MacBook Pro, making cross-machine compatibility a key consideration. The post explains why GPUs outperform CPUs for AI workloads, highlighting that matrix multiplication — the core operation in neural networks — benefits from the GPU's thousands of parallel cores. All inventory commands from the session are packaged into a reproducible shell script shared in a public companion repository.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Building Your Own CAPTCHA Solver Often Costs More Than Buying One

Teams automating web tasks frequently consider building in-house CAPTCHA solvers to avoid per-solve service fees, which typically run one to three dollars per thousand solves. However, a production-ready stack requires browser fleets, proxy management, fingerprint handling, concurrency controls, and ongoing monitoring — far beyond a simple script. Anti-bot vendors like Google and Cloudflare continuously update their systems, meaning any homegrown solution demands indefinite engineering maintenance estimated at 0.2 to 0.5 of a full-time engineer. Initial build time alone can reach one to three engineer-months, and residential proxy costs frequently exceed what a commercial solving service would charge. The analysis concludes that buying a solving service is usually more cost-effective unless CAPTCHA solving is a core business function justifying dedicated ownership.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Open-Sources Local AI Trading Research Workstation Built on Docker

A solo developer has released TradingSpy, an open-source AI-powered trading research workstation that runs entirely on Docker. The project was created after its developer grew frustrated with the infrastructure overhead of managing Jupyter notebooks and custom scripts for strategy backtesting. Commercial alternatives were ruled out due to high costs, proprietary cloud lock-in, and data privacy concerns around strategy harvesting. TradingSpy supports multiple LLM providers, market intelligence tools, and can be set up locally with three terminal commands. The developer has made the project freely available on GitHub, emphasizing it is intended for educational and research purposes only, not financial advice.

0 comments Read more at DEV Community