pr-monitor-and-manage-stop

name	pr-monitor-and-manage-stop
description	Clean-cancel companion to /pr-monitor-and-manage. Stops the PR-fleet Monitor tasks, clears monitoring state in session-state.json, and prints a final summary. Triggers on "/pr-monitor-and-manage-stop", "/pmm-stop", "stop monitoring PRs".
triggers	["pr-monitor-and-manage-stop","pmm-stop","stop monitoring PRs","stop PR fleet"]
argument-hint	(no arguments — stops the active PR-fleet Monitor in this thread)

Clean-cancel companion to /pr-monitor-and-manage. Use this to stop the PR-fleet manager without leaving a dangling main/auto-wake Monitor, stale monitoring state, or re-scan behind.

A stale pause marker left by a killed session is safely reconciled: the next /pr-monitor-and-manage invocation reads it and resumes (or the user runs /pmm-stop here to fully tear down), then re-runs discovery from scratch.

Resolve the state helper

Same three-candidate lookup as /babysit-pr-stop:

resolve_script() {
  local name="$1" candidate
  for candidate in \
    "$HOME/.claude/skills-worktree/.claude/scripts/$name" \
    "$HOME/.claude/scripts/$name" \
    ".claude/scripts/$name"; do
    if [[ -x "$candidate" ]]; then echo "$candidate"; return 0; fi
  done
  return 1
}
SESSION_STATE_SH=$(resolve_script session-state.sh) || { echo "ERROR: session-state.sh not found" >&2; exit 1; }

Step 1: Confirm there is something to stop

ACTIVE=$("$SESSION_STATE_SH" --get '.pmm_active' 2>/dev/null || echo null)
PAUSED_AT=$("$SESSION_STATE_SH" --get '.pmm.paused_at' 2>/dev/null || echo null)
STOP_PENDING=$("$SESSION_STATE_SH" --get '.pmm.stop_requested' 2>/dev/null || echo false)
MAIN_TASK_ID=$("$SESSION_STATE_SH" --get '.pmm_monitor_task_id' 2>/dev/null || echo null)
MAIN_GENERATION=$("$SESSION_STATE_SH" --get '.pmm_monitor_generation' 2>/dev/null || echo null)
AUTO_WAKE_TASK_ID=$("$SESSION_STATE_SH" --get '.pmm.auto_wake_monitor_task_id' 2>/dev/null || echo null)
AUTO_WAKE_GENERATION=$("$SESSION_STATE_SH" --get '.pmm.auto_wake_monitor_generation' 2>/dev/null || echo null)

• If $ACTIVE is true → proceed to teardown.
• If $PAUSED_AT is set (paused but not active) → proceed to teardown (clear pause marker + any re-scan Monitor).
• If both false/null, both IDs are null, and $STOP_PENDING != true → report "No active PR-fleet Monitor found." and stop.
• If $STOP_PENDING is true → continue the incomplete teardown even when the ordinary active/pause fields are absent.
• Any recorded ID → proceed to exact teardown even if the active/pause flags are stale.

Before calling TaskStop, atomically publish the cooperative stop guard:

"$SESSION_STATE_SH" \
  --set '.pmm.stop_requested=true' \
  --set '.pmm_active=false' \
  --set '.pmm_next_expected_tick_at=null'

The main skill's tick gate makes any already-emitted --tick event a no-op. Do not clear the stop marker until exact teardown is complete.

Step 2: Stop the main Monitor

Stop $MAIN_TASK_ID with exact TaskStop when present. When the captured $ACTIVE was true, a missing ID is degraded teardown, not success. Set MAIN_MONITOR_STOPPED=true only after successful TaskStop (or when the main task was not expected); otherwise set it to false.

When $STOP_PENDING was already true and an expected ID is missing (for example, an arm-publication failure), inspect the runtime task list. Clear the degraded state only after proving no matching PR fleet Monitor remains, or after stopping the exact task ID found there. Never treat a second invocation's now-false $ACTIVE value as proof that the previously unrecorded task disappeared.

Step 3: Cancel the auto-wake re-scan

If --auto-wake armed a re-scan at pause time, stop .pmm.auto_wake_monitor_task_id with TaskStop. Set AUTO_WAKE_MONITOR_STOPPED=true only after success; otherwise set it to false.

Step 4: Clear monitoring state (preserve everything else)

Immediately clear each ID and its matching generation whose exact TaskStop succeeded while preserving the stop marker and any pause marker/config/fleet. If either required stop failed or an active main task had no ID, retain the failed identity and .pmm.stop_requested=true, report PMM teardown incomplete, and stop here. A later /pmm-stop retries only the tasks that may still be live.

Only after every required stop succeeds, use one atomic terminal cleanup that preserves unrelated fields (other skills' state, PR tracking, budgets):

SET_ARGS=(
  --set '.pmm.stop_requested=false'
  --set '.pmm_active=false'
  --set '.pmm_next_expected_tick_at=null'
  --set '.pmm.paused_at=null'
  --set '.pmm.fleet_at_pause=null'
  --set '.pmm.config_at_pause=null'
  --set '.pmm_monitor_task_id=null'
  --set '.pmm_monitor_generation=null'
  --set '.pmm.auto_wake_monitor_task_id=null'
  --set '.pmm.auto_wake_monitor_generation=null'
)
"$SESSION_STATE_SH" "${SET_ARGS[@]}"

Leave pmm_in_flight, pmm_digest, pmm_digest_streak, and pmm_idle_streak in place as an audit trail — they are harmless once pmm_active=false, and a later /pr-monitor-and-manage re-invocation re-evaluates them against live PR state on its first tick. Do not touch cr_hourly, greptile_daily, prs, active_agents, or any non-pmm_* field. Note any PMM-owned entries still in active_agents (phase-a-fixer fix subagents) in the final summary — they may continue running until they exit on their own.

Step 5: Final summary

=== PR fleet monitoring stopped ===
Reason:   user /pmm-stop
Monitor:  stopped (no further ticks)
State:    pmm_active=false, stop/pause markers cleared
Auto-wake re-scan: <cancelled | none>
In-flight at stop: <list any pmm_in_flight PR # + skill, or "none">
Active subagents: <list any PMM-owned active_agents entries (match by pr + phase A, or task containing "PMM"), or "none">

If any PR had an in-flight phase-a-fixer subagent or /wrap dispatch when stopped, name it so the user knows that work was mid-flight. Re-run /pr-monitor-and-manage anytime to resume — it rediscovers the fleet from scratch and Step 2.5 aggregates any subagents that completed while monitoring was stopped.

Never print the successful summary when a required task was missing or failed to stop. In that case, name each retained task ID and say /pmm-stop must be retried before start/resume.