Our projects grow out of everyday work with AI agents. We’re interested in how people and agents share context, work through ideas, and learn from the process.
Human agency is part of that: understanding the work, contributing judgment, and shaping what happens next.
With support from OpenAI’s Codex for Open Source program.
Checks Claude Code, Codex, and Pi tool calls on macOS before they run. Blocks broad filesystem scans and recognizable credential reads, and tells the agent what to run instead.
Rust 1
Pi extensions for inline skill invocation, session IDs in the footer, Agent Guard checks, and a DeepSeek reasoning workaround. Each can be enabled on its own.
TypeScript 1
Claude Code plugins built on function hooks. preallow lets a model judge approve task steps in auto mode without the classifier.
Reopens Claude Code, Codex, Pi, and omp conversations in the panes Rex restores after its server stops, and carries worker results back to the orchestrator.
Search and count local Claude Code, Codex, and Pi session history in one SQLite index.
Checks Claude Code, Codex, and Pi tool calls on macOS before they run. Blocks broad filesystem scans and recognizable credential reads, and tells the agent what to run instead.
Shared Renovate preset for Entwining repositories, and the agent skill that explains it.
Pi extensions for inline skill invocation, session IDs in the footer, Agent Guard checks, and a DeepSeek reasoning workaround. Each can be enabled on its own.
Stores API keys, tokens, passwords, and SSH keys in iCloud Keychain and gives them to commands as environment variables, masking their values in output.
Loading…
Loading…
