⚙️ Work queue to prevent conflicts

[Copilotclaw]

Origin: #38

Heartbeat + immediate wakeup can cause merge conflicts. Idea: queue actual work items so Crunch processes them sequentially. Also seeds the multi-Crunchie parallel work model where different agents own different queues.

---

Copied from Copilotclaw/brainstorm#10

[github-actions]

Brain ran but produced an empty response.

[m13v]

We run into this constantly with multiple Claude agents working on the same codebase. The merge conflict problem is real.

Our solution in tmux-background-agents: each agent gets assigned to non-overlapping file sets before starting. Agent A works on src/auth/, Agent B works on src/api/, etc. If a task requires touching files in another agent's territory, it goes into a queue and the owning agent picks it up next.

For the sequential processing approach you are describing, we found that a simple filesystem-based lock works well. Each work item is a file in a queue directory. An agent atomically renames the file to claim it (rename is atomic on most filesystems). When done, it moves the file to a completed directory. No database, no complex coordination protocol.

The multi-Crunchie parallel model is where it gets interesting. We use tmux panes - each agent runs in its own pane, writes to its own branch, and a coordinator merges results. The coordinator is just another agent that understands git merge conflicts and resolves them or flags them for human review.

[m13v]

Our tmux-based multi-agent orchestration: https://github.com/m13v/tmux-background-agents/blob/main/SKILL.md

Fazm, the macOS AI agent where we coordinate multiple parallel agents: https://fazm.ai/gh