[Xehanort88]

Problem

A plugin that gives an enemy other skills or other gear has nothing to replace.

• Action patterns are read straight from the database entry in five places: Game_Enemy.makeActions, Game_Fight.chooseAction, the stage's skill preload, and two readers in the Tactical driver. The only way for a plugin to reach them is to swap enemy().actions on the database entry for the length of a call, and that still misses the Action and Tactical battle modes, which read the list outside any method a plugin can wrap.
• Enemy equipment (Enemy equipment, off by default #80) reads an enemy's authored rows and caches them in one method, rrEquipRows. A plugin that wants to equip an enemy from somewhere else has to wrap that method and rebuild the cached _rrEquips record by hand.

Change

Two methods on Game_Enemy, both returning the database lists by default, so a project without such a plugin plays as before.

• rrActionPatterns() returns the action patterns an enemy chooses from. All five readers go through it. The stage's preload keeps a fallback to enemy().actions for a battler that does not have the method.
• rrEquipSource() returns the rows an enemy's equipment is first read from. rrEquipRows copies what it returns, so the returned list is never changed, and a method that returns no list means no equipment. It is not read while the feature is off. The default slots 0.99.5 gives a never-edited enemy (one empty slot per equipment type) are built in this method.

Runtime revision 20261112.1, rebased onto 0.99.5. template/Demo/js is synced with sync-runtime.cjs.

Verification

• Two new tests in editor/tests/enemy-equipment.test.cjs, run on the shipped objects. One asserts that no runtime/reactor_*.js file reads enemy().actions beyond the stage's fallback.
• The whole editor suite was run on this branch and on main at 0.99.5, and the failing test names compared. One name differed each way, both unrelated to this change: a timer test in memory-budget.test.cjs failed only in the run with the change and passed three times in a row when run alone, and a character-builder test failed only on main.
• In a real game with a plugin that overrides both methods: a battle test under the turn-based mode ran several turns with no errors, and Game_Fight.prototype.chooseAction called on a live enemy picked the skills the plugin supplied.

Limits

• The Tactical driver's two readers are covered by the source assertion only. No Tactical battle was played.
• chooseAction was called directly on a live enemy; no Action-mode battle was played end to end.
• No editor change, so nothing here was checked on screen in the editor.
• The real-game run was made before the rebase, on the 0.99.4 runtime with the same two methods. After the rebase only the tests were run.

🤖 Generated with Claude Code