One shared repo to rule Claude, Cursor, and the rest of your AI agents

What it does

TeamAI is a TypeScript CLI that treats a shared repository as a control plane for AI development tools. Administrators maintain skills, rules, hooks, and MCP server definitions in the repo, and the CLI distributes those resources into the native configuration directories of Claude Code, Cursor, Codex, and several other agents. The tool also attempts to capture hard-won team knowledge by detecting high-friction AI sessions and prompting users to share what they learned.

The interesting bit

The clever angle is the friction-based session scoring: a Stop hook measures signals like interrupted tasks, denied tool calls, and retries to identify sessions that actually taught someone something, then nudges the user to contribute a summary back to the shared repo. It is essentially GitOps for agent configuration, complete with branch-and-merge review workflows and role-based filtering so different members receive only the skills they need.

Key highlights

• Supports at least ten AI agents—including Claude Code, Cursor, Codex, and DeepSeek Harness—with varying degrees of feature coverage.
• Distributes MCP server definitions, custom hooks, environment templates, and even npm packages from a central repository to each tool’s native config paths.
• Lets teams subscribe to external skill repositories and merge them into their own harness.
• Changes are reviewed before merging, and teammates automatically synchronize on their next session start.
• Includes role and tag filters so large teams can avoid cluttering every workstation with irrelevant skills.

Caveats

• Feature parity across agents is uneven; the README’s own compatibility matrix shows that OpenCode, Hermes, and DeepSeek Harness lack several execution and improvement features.
• The “Team Context” and “Team Improvement” sections are cut off mid-sentence in the README, so the full behavior of knowledge recall and dashboards remains unclear from the provided sources.

Verdict

Worth a look if your team is already juggling multiple AI coding tools and wants a single source of truth for shared prompts and rules. Skip it if you are a solo developer using one editor, or if your agents’ config formats are too idiosyncratic to fit a shared schema.