Skill Generation is the current bridge between semantic data and moment-to-moment combat. The goal is not to attach a random effect to every card. A generated skill should remain recognizable as an expression of the work that produced it.
Meaning enters a controlled gameplay vocabulary
Embedding similarity can surface concepts, but raw concepts are not balanced game effects. The pipeline therefore separates semantic selection from effect construction. Concepts choose from controlled trigger, target, cost, cooldown, and effect families; balance rules constrain the resulting values.
The system can use an active/passive distinction when it changes player understanding, but it does not create two systems merely for symmetry. Every additional category must produce a real deck-building decision.
Assignment is part of preparation
Generated skills appear in Battle Prep with their Trigger, Cost, Cooldown, and Effect. They can be assigned or removed before combat. The UI prevents uncontrolled duplication by checking the selected deck card, the card’s current assignments, and whether the same skill_id is already used elsewhere in the deck.
Generation creates candidates. The player’s allocation decision creates the deck.
Fallbacks must remain visible
If no generated skill matches the current deck filter, the interface can show all available results rather than presenting an unexplained empty panel. This is a usability fallback, not permission to ignore semantic relevance.
Current implementation status
Card and skill allocation UI, removal, saved composition, and deck-level duplicate checks are in place. The remaining work is refining the concept-to-effect mapping, balancing values, and ensuring the final battle can execute every generated effect consistently.
Why this matters
ThoughtMap already answers which works are conceptually near one another. Skill Generation asks a harder question: how should that meaning behave? Solving it is what turns the research database from a card source into a game system.