Show HN: Crew – Let Claude Code agents talk to each other

What Happened

A developer has created an open-source tool called Crew that enables multiple Claude Code sessions running on the same repository to communicate with each other. Instead of manually juggling separate worktrees or relying on external coordination, Crew injects contextual information—status updates, recaps, and recent transcript entries—from every active session into each other’s context window. This allows Claude Code agents to effectively “see” what their peers are doing and, crucially, to message one another directly.

The tool hooks into Claude Code’s existing architecture, meaning it doesn’t require a separate infrastructure or API calls. It operates as a lightweight overlay that synchronizes state across concurrent sessions.

Why It Matters

This development addresses a practical pain point for developers who run multiple Claude Code instances on the same codebase—a workflow that is increasingly common as AI-assisted coding becomes more collaborative. Without such coordination, each session operates in isolation, unaware of changes made by another instance. This can lead to conflicts, redundant work, or context loss when switching between tasks.

Crew’s approach is notable for its simplicity: it leverages Claude Code’s own context injection mechanism rather than building a new communication protocol. This means the AI agents remain the primary actors, but now with shared awareness. The ability for agents to message each other introduces a primitive form of multi-agent collaboration within a single repository—something that previously required manual orchestration or external tools like Slack bots or custom scripts.

For AI practitioners, this is a step toward more autonomous, parallelized coding workflows. Instead of a single Claude Code session handling everything sequentially, teams of agents can divide work—one refactoring a module, another writing tests, a third reviewing output—while staying synchronized. This mirrors how human developers already work in parallel, but with the advantage that AI agents can share raw context instantly.

Implications for AI Practitioners

First, Crew lowers the barrier to running multi-agent coding sessions. Developers no longer need to design complex coordination logic; the tool handles context sharing transparently. This could accelerate adoption of parallel AI coding in production environments.

Second, it raises questions about context window management. Injecting transcripts from multiple sessions into each agent’s context consumes tokens. Practitioners will need to consider how many concurrent sessions are practical before hitting context limits or degrading response quality.

Third, Crew highlights a growing trend: treating AI agents as collaborative peers rather than isolated tools. As more developers run multiple Claude Code sessions, tools like Crew will become essential infrastructure. This also opens the door to more sophisticated patterns—for example, one agent acting as a supervisor that delegates subtasks to others.

Finally, because Crew is open-source and community-driven, it invites experimentation. Developers can extend it to add features like task prioritization, conflict resolution, or logging. This aligns with the broader movement toward agentic workflows where AI systems coordinate autonomously.

Key Takeaways

• Crew enables multiple Claude Code sessions on the same repo to share context and message each other, solving a real coordination problem for parallel AI coding.
• The tool works by injecting session status and transcripts into each agent’s context, requiring no external infrastructure.
• For practitioners, this makes multi-agent coding more practical but introduces trade-offs around context window usage and session management.
• Crew reflects a shift toward treating AI agents as collaborative peers, with potential for further community-driven enhancements in autonomous coordination.