diff --git a/scripts/deploy-stack.sh b/scripts/deploy-stack.sh
index 099b0fc..7247ea9 100755
--- a/scripts/deploy-stack.sh
+++ b/scripts/deploy-stack.sh
@@ -131,7 +131,13 @@ if [ -x "$OPS_LOG" ] && [ "${DEPLOY_NO_CLAIM:-0}" != 1 ]; then
   "$OPS_LOG" claim "$HOST" "$STACK" --ttl "${DEPLOY_CLAIM_TTL:-30m}" \
     --why "deploy-stack.sh $HOST $STACK" -q || claim_rc=$?
   case "$claim_rc" in
-    0) CLAIMED=1 ;;
+    0)  CLAIMED=1 ;;
+    10) # We only refreshed a claim this agent already held for a wider
+        # operation. Do NOT release it on the way out — dropping someone's
+        # multi-step claim mid-operation is exactly the exposure the claim
+        # exists to prevent.
+        CLAIMED=0
+        echo "note: $HOST/$STACK was already claimed by you — leaving that claim in place." ;;
     3) echo "error: refused — see the claim above. Wait for the holder, coordinate" >&2
        echo "       on althing, or override with DEPLOY_NO_CLAIM=1 if it is dead." >&2
        exit 3 ;;
diff --git a/scripts/ops-log b/scripts/ops-log
index 932281f..e4292fd 100755
--- a/scripts/ops-log
+++ b/scripts/ops-log
@@ -54,6 +54,8 @@ EXIT CODES
   0  success / claim acquired / nothing to report
   2  usage error
   3  claim refused (held by another agent)
+  10 claim REFRESHED — you already held it. The caller did NOT acquire it and
+     must NOT release it on the way out (see deploy-stack.sh).
   4  audit found unlogged changes
   5  audit could not reach every host (incomplete — NOT the same as clean)
 """
@@ -246,6 +248,17 @@ def cmd_claim(args) -> int:
         _drop_claim(args.host, args.target)
         record(args.host, "claim-broken", args.target,
                outcome="changed" if state == "theirs" else "ok", detail=detail)
+    if state == "mine":
+        # Already ours: leave the holder file ALONE. A sub-tool refreshing a
+        # claim would otherwise overwrite the reason and TTL the original
+        # claimant chose — so a 45-minute "rollout in progress" becomes
+        # "deploy-stack.sh <host> <stack>", and whoever gets refused reads the
+        # wrong story about why. Report 10 and change nothing.
+        if not args.quiet:
+            print(f"already claimed by you: {args.host}/{args.target} "
+                  f"({(holder or {}).get('why') or 'no reason given'}, "
+                  f"{holder_age(holder)})")
+        return 10
     d = CLAIMS_DIR / claim_key(args.host, args.target)
     try:
         d.mkdir(parents=True, exist_ok=False)   # atomic: this IS the acquire
@@ -269,6 +282,12 @@ def cmd_claim(args) -> int:
     if not args.quiet:
         print(f"claimed {args.host}/{args.target} for {args.ttl} "
               f"({holder['why'] or 'no reason given'})")
+    # Exit 10 when this only REFRESHED a claim the caller already held. A tool
+    # that claims-then-releases around its own work must not drop a longer
+    # claim wrapping a multi-step operation. Measured 2026-09-19: a 45-minute
+    # operation claim on nh3-docker/althing-post-office was silently released
+    # by deploy-stack.sh's exit trap partway through the rollout it was
+    # protecting.
     return 0