long-waitlisted

# Long Wait Run `scripts/long_wait.py` by absolute path from the task's working directory; do not change into this skill directory first. Relative workload paths inherit that working directory. Primary agent only. Requires Linux or macOS. Native Windows is unsupported. ```bash LONG_WAIT_SCRIPT=/absolute/path/to/long-wait/scripts/long_wait.py python3 "$LONG_WAIT_SCRIPT" probe python3 "$LONG_WAIT_SCRIPT" after 6h python3 "$LONG_WAIT_SCRIPT" run --description "Train baseline model" --timeout 2d -- command arg python3 "$LONG_WAIT_SCRIPT" until --timeout 2d --interval 1m -- predicate arg ``` Choose a mode by ownership and durability: - `probe`: two-turn barrier. `state: delivered` means queue accepted marker, not probe confirmed. After command succeeds, end turn immediately. Do not register real wait or do other work. Matching `[long-wait-probe:v1]` in next turn confirms route; only then register real wait. - `after DURATION`: use when only elapsed time matters; accepts `30s`, `10m`, `6h`, or `2d`. - `run -- COMMAND...`: own cheap, restartable, resumable, or safely terminated work. Wake on exit, failure, timeout, or exhausted retries. Retry only idempotent work; `--max-retries` defaults to `0`. Timeout sends `SIGTERM` to the process group, then `SIGKILL` after two seconds. Partial side effects and deliberately detached descendants may remain. - `until --timeout DURATION -- COMMAND...`: observe expensive, non-resumable, or independently supervised work through