Overview

Shared Agents gives your team one Git-synced home for AI assistant knowledge: skills (workflows), rules (instructions), and learnings (institutional memory). It works across Cursor, Claude Code, Zed, Codex, and other tools without locking you to a single vendor.

The problem

Teams use many AI tools. Each session starts cold — no shared workflows, no memory of past bugs or decisions, and knowledge trapped in chats or private notes.

The solution

After bootstrap, teammates run sa sync to pull Core + team content and link skills/rules into each IDE.

Core concepts

Skills

Repeatable agent workflows (Markdown under skills/). Example: sa-cli, capture-learning, shopware-workflow.

• Core skills — public, contributed via PR
• Team skills — private, sa skill new → team/skills/<name>/SKILL.md

Rules

Shared instructions per tool. Cursor uses .mdc symlinks; other tools get merged blocks in AGENTS.md / CLAUDE.md.

• Core rules — rules/*.mdc
• Team rules — flat team/rules/*.mdc (no review workflow)

Learnings

What the team already figured out (bugs, stack quirks, decisions). Human-in-the-loop:

1. Agent drafts → team/learnings/pending/ (after you say yes)
2. Human approves → sa review → approved/
3. Agents read only approved learnings

See Learnings.

Architecture

  GitHub (Core)              Your private remote (Team)
        │                              │
        └────────── sa sync ───────────┘
                        │
                        ▼
              ~/.shared-agents  ──►  Cursor · Claude Code · Zed · …

Repository layout

shared-agents/
├── skills/           # Core skills
├── rules/            # Core rules
├── adapters/         # Per-tool wiring (manifest.json)
├── scripts/          # sa CLI, sync, bootstrap
├── docs/             # This documentation (website + repo)
└── team/             # Private data (separate Git remote)
    ├── learnings/    # pending/ + approved/
    ├── rules/        # flat *.mdc
    └── skills/

Quick commands

Next steps

1. Install Shared Agents
2. Set up your team repo
3. Configure adapters for your IDEs
4. Contributing if you improve Core