better-goal

{
  "name": "better-goal",
  "description": "Hold a long autonomous run to a finish line using command hooks and a stall watcher this skill creates itself, rather than the built-in /goal. /goal is a prompt Stop hook judged by a small model on transcript text alone, so it grades what the run said rather than what is true, and Claude Code overrides any Stop hook after 8 consecutive blocks while reporting that turn as completed. better-goal (formerly goal-harness) grounds the finish line in the repo's real worklist, writes the brief to docs/goals/goal-<slug>.md, turns \"done\" into gate commands judged by exit code, preflights the settings that silently end a run, and registers four events rather than one: Stop runs the gates, StopFailure records a turn that ended on an API error, SessionEnd records a session that closed while armed, and a SessionStart sentinel tells the next session what is still armed and cold. It proves the hook actually loaded instead of assuming it, because a hook registered mid-session into a .claude/ Claude Code is not watching is written correctly and never fires. Its watcher reads the session transcript, so a long turn is not mistaken for a dead run. It stops deliberately when a gate fails identically turn after turn rather than re-sending the same failure, and answers \"has it been met?\" and \"why did you stop?\" from a ledger instead of a guess.",
  "version": "2.2.0",
  "author": {
    "name": "Fledgeling",
    "email": "[email protected]"
  }
}