Claude Code

Tell the agent to acquire before it starts and release when it finishes, for example in a prompt, a slash command, or CLAUDE.md:

Before any other work, run `waitlet acquire <issue-id>` as a background command
(run_in_background) and wait for it to finish; it exits once a slot is free.
If it exits with 3, another agent already has this issue: stop.
When you are done, or if you stop early, run `waitlet release`.

A background command doesn't hit Claude Code's 10-minute tool timeout, and the agent spends no tokens while it waits: Claude Code wakes it when the command exits. Where background commands aren't available, use waitlet acquire <id> --timeout 9m in a loop: exit code 75 means "run it again", and the queue place is kept for 60 seconds between attempts.

Hooks

Optional hooks in .claude/settings.json (or ~/.claude/settings.json) make the TTL an idle timeout and free the slot when the session ends:

{
  "hooks": {
    "PostToolUse": [{ "hooks": [{ "type": "command", "command": "waitlet renew -q || true" }] }],
    "SessionEnd": [{ "hooks": [{ "type": "command", "command": "waitlet release -q || true" }] }]
  }
}

Then acquire with a short TTL, such as waitlet acquire <id> --ttl 20m: every tool call renews the lease, and a session that stops calling tools for 20 minutes gives its slot up.

Hooks find the session through their parent processes. If that fails (for example, Claude Code running under a renamed binary), pass --pid "$PPID".