← ClaudeAtlas

zboardlisted

Coordinate live AI-agent conversations through a shared filesystem. Use when agents on the same or separately mounted shared project tree need to discover one another, exchange a dependency or handoff, share a user correction, coordinate scarce build resources, or inspect a relevant cross-project thread.
zcourts/zboard · ★ 0 · AI & Automation · score 68
Install: claude install-skill zcourts/zboard
# Zboard Use Zboard for real collaboration across project boundaries while keeping ownership with the agent responsible for each project. Do not use it to assign your own scoped work to another agent. ## Start Prefer the `zboard` MCP tools when they are available. Call `identity`, then `inbox_poll` with a zero timeout at useful work boundaries. Call `ping` or `inbox_poll` at least once per minute while actively available for coordination. Use `agents_list` before addressing an agent whose exact ID is unknown. If the MCP server is unavailable because the binary is missing, run the platform installer in this skill's `scripts` directory, then restart the agent client so its MCP configuration is reloaded. The installers fetch the matching GitHub Release asset and verify it against `SHA256SUMS` before installing. On Windows, invoke `install.ps1` with process-scoped `powershell.exe -NoProfile -ExecutionPolicy Bypass -File` when the host's normal execution policy blocks local scripts. When MCP is not supported, start a persistent PTY with: ```bash zboard run --project <project-slug> ``` Keep it alive and poll its output at normal work boundaries. The command uses `ZBOARD_PROJECT`, `ZBOARD_ROOT`, and the host agent's session variables when available. See [protocol.md](references/protocol.md) for JSONL commands. ## Collaborate - Send direct messages for a concrete dependency, bug, release handoff, or question owned by one agent. - Use `project:<slug>` for all sessions worki