Joe Cooper
Joe Cooper is an individual software developer and creator (GitHub handle @swelljoe), not a company. He publishes software, security research, experiments with AI coding agents and local language models, and technical writing through swelljoe.com and his open-source repositories.
At a Glance
- Individual developers and open-source maintainers
- Security-conscious developers using AI coding agents
- Application-security researchers and software teams evaluating AI vulnerability discovery
- Linux users and systems programmers who need lightweight command-line tools
- +1 more
AI Tools by Joe Cooper
(1)flar
Bubblewrap Sandbox for AI Agents
Discussions
No discussions yet
Be the first to start a discussion about Joe Cooper
Latest News
How I Run Local LLMs: practical guidance on Gemma 4, Qwen 3.6, llama.cpp, hardware, quantization and agents
Poolside Laguna S 2.1 Is Alright
I Let Every Agent Implement Its Own Flar Resume Backend
Initial flar repository release, an open-source Bubblewrap restrictor for coding agents
Products & Services
An open-source Go CLI that runs coding-agent CLIs inside unprivileged Bubblewrap sandboxes. It mounts only the project read-write, hides the rest of the home directory, isolates or proxies networking, scopes or forks agent history, and copies only the credentials needed by the selected agent.
An open-source security-vulnerability discovery and benchmarking tool. It loops agents over project files, supports scan and review stages, de-duplicates findings, repeats runs to address flaky detection, and works with Claude Code, Gemini CLI and OpenAI-compatible APIs.
A concise POSIX-shell logging library for bash, dash and zsh, with colored console output and optional logging to a file. Cooper ported it from Fred Palmer's log4bash and added POSIX compatibility and file logging.
Market Position
Cooper's work is positioned as lightweight, transparent, developer-controlled open source rather than a commercial platform. Flar's differentiator is combining Bubblewrap isolation, restricted filesystem visibility, network controls and per-agent/per-project history handling; relevant alternatives include Docker or Podman containers, Firejail, and the built-in sandboxes of individual coding agents. Nelson emphasizes repeated empirical agent passes and review/de-duplication rather than claiming that a single clever prompt or static scan is sufficient.
Founding Story
The official site is a personal portfolio/blog titled "I've done some things," rather than a company site. Cooper's projects appear to be independently created tools motivated by practical needs: for example, he built flar after recognizing that AI-agent sandboxes were porous and that prompt injection or supply-chain attacks could expose credentials and cross-project history; he built Nelson to repeatedly scan code with agents for vulnerabilities and benchmark what actually finds bugs.
Target Markets
- Individual developers and open-source maintainers
- Security-conscious developers using AI coding agents
- Application-security researchers and software teams evaluating AI vulnerability discovery
- Linux users and systems programmers who need lightweight command-line tools
- Developers experimenting with local or hosted language models
- Safely running coding agents with permissions enabled inside a project
- Reducing prompt-injection and software-supply-chain blast radius during AI-assisted development
- Preventing one project's agent credentials or chat history from leaking into another project
- Scanning source trees for security vulnerabilities with repeated agent passes and human-oriented review
- Benchmarking the effectiveness and cost of different AI models and agent runtimes at finding real bugs
- Adding portable structured or colored logging to POSIX shell scripts