Why markdown plans break when you run multiple AI agents

Описание и применение

Многопользовательская координация AI‑агентов через общий markdown‑файл приводит к конфликтам и потере истории. Решение – сервер MCP (workstream‑mcp), который обеспечивает атомарные заявки, истечение срока, проверку переходов, ревью и неизменяемую историю.

Задача

Coordinating multiple AI coding agents via a shared markdown checklist leads to conflicts, stale claims, and lost history.

Результат

Using workstream‑mcp eliminates duplicate task claims, ensures timely claim expiry, enforces legal state transitions, provides review workflow, preserves immutable history, and offers real‑time visibility, enabling reliable multi‑agent coordination.

Рабочий процесс

• 1. Agent reads plan via MCP tool call
• 2. MCP claims next task using SELECT … FOR UPDATE SKIP LOCKED within a transaction
• 3. Agent performs work and updates task status
• 4. MCP enforces legal transitions and dependency rules
• 5. Review step: developer agent submits, reviewer approves or rejects
• 6. MCP records every change as an event in an append‑only log
• 7. Notifications sent through an outbox to GitHub/Slack
• 8. Dashboard displays current claims, expiries, and dependencies

Доказательства

Most of us coordinate AI coding agents the same way: a markdown file in the repo with a checklist.

Источник

Two agents pick the same item.

Источник

Claims run in one Postgres transaction with SELECT … FOR UPDATE SKIP LOCKED.

Источник

A test suite runs exactly that.

Источник

Every change writes an event. Database triggers block updates and deletes on the event log, so the history can't be rewritten after the fact.

Источник