/operator-recall — Halt-Class Condition Detector + Pager

Canonical surface: `scripts/dispatch/operator-recall.sh` (INFRA-626). Any harness invokes the same script.

When to invoke this skill

• Diagnostic mode (--check-only) — silent check: "is any halt-class condition active right now?" Exit 1 if yes (with a reason printed). No side effects. This is the default for agent use.
• Active page (no flag, or --condition X --reason Y) — emits ambient kind=operator_recall AND POSTs to CHUMP_OPERATOR_RECALL_URL if set. Use sparingly — this pages a human.

Routing

Arguments passed: $ARGUMENTS. Common forms:

# Diagnostic — check, exit 1 if any condition active (most common for agent use)
scripts/dispatch/operator-recall.sh --check-only

# Active page — operator-initiated, explicit
scripts/dispatch/operator-recall.sh --condition AUTH_DEAD --reason "fleet_auth_storm cluster: 5 workers exited in last hour"

# Auto-detect all conditions, emit + notify any that fire
scripts/dispatch/operator-recall.sh

scripts/dispatch/operator-recall.sh $ARGUMENTS

If $ARGUMENTS is empty, default to --check-only (the safe diagnostic mode). Confirm with the user before sending an active page.

The four halt-class conditions

Behavior rules

• Never call the active-page mode without explicit user instruction. --check-only for any agent-initiated diagnostic.
• If a condition fires in --check-only mode, surface the condition name + reason to the user and propose next steps (page operator? auto-mitigate via a script? file a gap?).
• The recall emit is idempotent — cooldown-gated by CHUMP_OPERATOR_RECALL_COOLDOWN_SECS (default 1800) — so re-triggering on the same condition within 30 min won't double-page.

When NOT to use this

• For non-halt-class issues (slow PR, single flake, one stuck check) — use /fleet-doctor or direct script invocation
• For status summaries — use /fleet-brief
• For "missing feature" investigation — use /verify-existence