Repository navigation
bulkhead: keeping parallel card agents from stopping each other's dev servers #3502
dooly6276
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Disclosure: I maintain bulkhead, an MIT licensed set of PreToolUse hooks for coding agents.
When I start several cards at once, each attempt gets its own worktree, but all of the agents run on one machine. That is where they meet: one agent restarts its server with
pkill nodeand the dev servers of the other cards go down, two attempts at the same app both want port 3000, or three cards start a build, a test run and Playwright in the same minute and the machine stalls.bulkhead checks each shell command before the agent runs it. kill-guard blocks a kill by name that reaches another session's processes, port-guard blocks starting a server on a port another agent session holds and names a free one, and memory-guard blocks heavy commands when free RAM is low. The agent gets the reason and a safer command back, so it can carry on. It loads in Claude Code, Codex, Gemini CLI, Cursor and opencode, which are among the agents a card can run.
https://github.com/dooly6276/claude-bulkhead
All reactions