Finding
Long-running tests, servers, and watchers become operationally weak when they are started as unmanaged terminal work instead of tracked background processes with polling, readiness checks, and cleanup.
Current
A real Hermes installation often needs to run commands that do not finish immediately: test suites, dev servers, file watchers, build processes, crawlers, or health checks. The weak point is usually not starting the process; it is controlling the lifecycle afterward. Without a standard background-process habit, agents may wait unnecessarily, lose output, forget cleanup, run duplicate watchers, or treat a long-running server as complete before verifying readiness.
Suggested
- Add a background-process decision rule for long-running commands. Exact change: add a “Background process rule” to
SOUL.mdor the main operator runbook: “Use tracked background processes for servers, watchers, and long tests; use foreground only for commands expected to complete quickly; never use shell-level&,nohup, ordisownwhen Hermes process tracking is available.” - Create a lifecycle checklist for every background run. Exact change: add a runbook section named “Background process lifecycle” requiring: start with a clear purpose, capture the process/session handle, poll or wait for completion, inspect logs when needed, verify readiness with a health check or expected output, and kill/cleanup the process when it is no longer needed.
- Add a completion verification habit to testing and server workflows. Exact change: patch the testing or dev-server skill with: “Before final response, confirm whether each background process is still needed, stopped, or intentionally left running; include the verification used, such as passing tests, a reachable health endpoint, or a clean process exit.”
Impact
This improves control over long-running work without promoting background processes into a strategy for everything. Hermes can keep working while tests, servers, or watchers run, but still preserve visibility into logs, failures, readiness, and cleanup. It reduces wasted time, orphaned processes, duplicate servers, and false success reports from tasks that only appeared to start correctly.
Effort
Small — the change is mostly a runbook and skill discipline update, plus a habit of polling, verifying, and cleaning up tracked processes during long-running work.
Public page note
Safe public content includes the maturity principle, generic examples of long-running commands, lifecycle rules, readiness checks, and cleanup habits. Internal-only content includes real process IDs, raw logs, private test output, local ports tied to sensitive services, filesystem paths, credentials, environment values, and any live operational control details.