diff --git a/hooks-handlers/post-tool-use.py b/hooks-handlers/post-tool-use.py
index 4fa77b9..b607c46 100755
--- a/hooks-handlers/post-tool-use.py
+++ b/hooks-handlers/post-tool-use.py
@@ -457,8 +457,13 @@ def main():
             # resolving encounters before the user can react.
             if output and not encounter_still_present(existing, output, catalog):
                 if existing.get("catch_pending"):
-                    # User invoked /buddymon catch — hold the monster for them
-                    pass
+                    # User invoked /buddymon catch — hold the monster for this run.
+                    # Clear the flag now so the NEXT clean run resumes normal behavior.
+                    # The skill sets it again at the start of each /buddymon catch call.
+                    existing["catch_pending"] = False
+                    enc_data = load_json(BUDDYMON_DIR / "encounters.json")
+                    enc_data["active_encounter"] = existing
+                    save_json(BUDDYMON_DIR / "encounters.json", enc_data)
                 elif existing.get("wounded"):
                     # Wounded: 35% chance to flee per clean run (avg ~3 runs to escape)
                     if random.random() < 0.35:
diff --git a/skills/buddymon/SKILL.md b/skills/buddymon/SKILL.md
index e65887c..4faacb4 100644
--- a/skills/buddymon/SKILL.md
+++ b/skills/buddymon/SKILL.md
@@ -298,8 +298,9 @@ if not buddy_id:
 
 enc = encounters.get("active_encounter")
 
-# Clear catch_pending before rolling (win or lose)
-enc["catch_pending"] = False
+# catch_pending is cleared by the PostToolUse hook after it fires on this
+# Bash run — do NOT clear it here or the hook will see it already gone and
+# may auto-resolve the encounter on the same run as the catch attempt.
 
 buddy_data = (catalog.get("buddymon", {}).get(buddy_id)
               or catalog.get("evolutions", {}).get(buddy_id) or {})
@@ -339,7 +340,8 @@ if success:
     json.dump(encounters, open(enc_file, "w"), indent=2)
     print(f"caught:{xp}")
 else:
-    # Save cleared catch_pending back on failure
+    # Leave catch_pending as-is — the PostToolUse hook clears it after this
+    # Bash run completes, giving one full clean run before auto-resolve resumes.
     encounters["active_encounter"] = enc
     json.dump(encounters, open(enc_file, "w"), indent=2)
     print(f"failed:{int(catch_rate * 100)}")