cross-machine-coordination

Create a YAML file in .squad/cross-machine/tasks/ with the task id, source_machine, target_machine, payload command, and status pending, then commit and push to git. The target machine's agent pulls it on its next watch cycle, validates it, executes it, and pushes a result file.

Create a GitHub issue with the label squad:machine-{name} for the target machine, describing the task in the issue body. The remote agent detects the labeled issue, executes the task, comments with the result, and closes the issue.

Tasks are YAML files requiring id, source_machine, target_machine, priority, created_at, task_type, payload with command and resources, and status. Results are separate YAML files containing exit_code, stdout, stderr, duration_seconds, and artifact paths.

Every task passes schema validation, a command whitelist, path validation, and resource limits before execution. Tasks run as an unprivileged user with timeouts and memory caps, and all executions are logged immutably in git history.

Failed tasks produce a result file with status failed, the exit code, and captured stderr, committed to git for audit. Timed-out tasks are killed and marked status timeout; the source machine can retry by re-pushing the task as pending.

Execution latency depends on the agent's poll interval of 5-10 minutes, so it is unsuitable for real-time work. Only whitelisted commands can run, and network outages delay execution until connectivity is restored, though no queued tasks are lost.