Tencent/teamai-cli

TeamAI CLI – A shared‑repo manager for AI‑assistant tools

What it is

• A command‑line tool (distributed via npm as teamai-cli) that lets a software team keep the skills, rules, docs, environment settings, hooks, and other resources used by a variety of AI coding assistants (Claude Code, Codex, Cursor, Qoder, CodeBuddy, OpenCode, WorkBuddy, etc.) in sync across all members.
• The resources live in a shared Git repository (GitHub, GitLab, GitCode, CNB, TGit, or any private Git service). Team members push changes as merge requests; once merged, a teamai pull automatically injects the updates into each local AI tool.

Core concepts

Layer	Goal	CLI support
Team Execution	Make every agent follow the team’s conventions (skills, rules, docs, env, hooks, MCP servers)	init, pull, push, install, hooks, mcp, source
Team Context	Give agents a shared knowledge base (learned sessions, code‑base graph, team wiki)	recall, import, codebase --lint
Team Improvement	Turn friction‑rich sessions into team‑wide learnings and dashboards	session save, digest, dashboard, recall promote/maintenance

Key features

• Distributed skill/hook repo – store Claude, Codex, Cursor, etc. skill files under ~/.<agent>/skills/ and keep them version‑controlled.
• Role‑ and tag‑based filtering – admins define which skills each role or tag should receive (teamai roles, teamai tags).
• Source subscription – pull in additional skill repos from other teams or a shared org repo (teamai source add …).
• MCP (model‑control‑plane) server config – declare remote inference endpoints once and have them written into each tool’s native config.
• Automatic session‑learning – when a session generates “friction” (interrupts, tool retries, corrections), the CLI can suggest summarising the experience and pushing it to the repo.
• Team knowledge graph – teamai import parses source code (AST via Tree‑Sitter for TS/JS/Python/Go, heuristic regex for other languages) into a graph stored under teamwiki/. The graph is used to re‑rank recall results.
• Recall sub‑agent – optional plug‑in that lets any supported AI tool query the shared knowledge base before answering a task (teamai recall enable).
• Dashboards & digests – weekly usage reports, live session status, and KB‑health pages help teams monitor token consumption, intervention rates, and knowledge‑base coverage.
• CI integration – teamai ci extract-mr can automatically pull knowledge from merge‑request discussions and post comments.

Typical workflow

1. Create a shared repo (or use a template from the teamai-hub org).
2. Run teamai init <repo‑url> – logs in, registers the user, injects the initial hooks.
3. Edit or add skills/rules locally, then teamai push → opens a merge request for review.
4. After the MR is merged, every member’s next AI session triggers a teamai pull (via a SessionStart hook) that syncs the latest resources into their local AI tools.
5. When a session is “friction‑rich”, run /teamai-share-learnings (or teamai session save) to capture a summary that becomes a new knowledge entry.
6. Periodically run teamai digest or open teamai dashboard to see team‑wide metrics and KB health.

Installation

npm install -g teamai-cli   # requires Node.js

After installation, the CLI is ready to run teamai init ….

Supported AI agents (as of the README) – Claude Code, Codex, Cursor, Qoder, CodeBuddy, OpenCode, WorkBuddy, OpenClaw, Hermes, DeepSeek Harness. The table in the README shows which capabilities (skills, rules, docs, env, agents, hooks, MCP) are currently implemented for each.

License

• MIT (see LICENSE).

Who might use it

• Engineering teams that rely heavily on AI coding assistants and want a single source of truth for prompts, plugins, and policies.
• Organizations that need role‑based distribution of AI‑generated code‑review or debugging rules.
• Teams looking to capture and reuse “tribal knowledge” from AI‑assisted debugging sessions.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.