Card Game Agent Skill

Card Game

A playbook for card games — card data, the deck/hand/discard zones, the turn structure, and

how card effects resolve. This is a compositional skill: it models cards as data and wires

them to UI. It does not re-teach data assets or UI nodes; it defines the zone model, the draw

machinery, and the effect-resolution rules that keep a card game correct and bug-free.

When to use

• Use when building any game where the core objects are cards moving between zones

(deck → hand → play → discard): deckbuilder, TCG/CCG, solitaire, roguelike deckbuilder.

• Use when designing draw/shuffle/reshuffle, turn structure, card costs, or how effects resolve.

When *not* to use: board/tile state with matching rules → puzzle. RPG with an

incidental card battler → start from rpg. For defining cards as assets, use godot-resources

/ unity-scriptableobjects; for the hand/drag UI, use godot-ui-control.

Core loop

**Draw to your hand → spend resources to play cards → effects resolve and change the board →

end the turn (cleanup/discard) → opponent/next phase → repeat until a win condition.** Depth

comes from the *combinations* a hand allows; the engine's job is to resolve them unambiguously.

Must-have systems

• Card data — id, name, cost, type, text, and an effect spec (data, not code).
• Zones — deck (draw pile), hand, play/board, discard, exile/removed; cards live in exactly one.
• Draw + shuffle + reshuffle — draw from deck to hand; reshuffle discard into deck when empty.
• Turn structure — phases (untap/draw/main/combat/end) as a state machine.
• Resource system — mana/energy/actions that gate how much you do per turn.
• Effect resolution — apply a card's effects in a defined order; handle targets and triggers.
• Win/loss condition — life total, deck-out, objective.
• UI — hand layout, drag/drop or tap-to-play, zone counts, targeting affordances.

Design knobs

| Knob | Effect | Notes |

|------|--------|-------|

| Starting hand / draw-per-turn | tempo, consistency | More draw = less variance. |

| Hand size limit | hoarding vs. use | Discard down at end of turn. |

| Deck size (min) | consistency | Smaller = more reliable combos. |

| Resource curve | what's playable when | "Mana curve" paces power. |

| Card rarity / power budget | balance | Stronger cards cost more / are rarer. |

| Determinism vs. randomness | skill vs. swing | Shuffle + random effects add variance. |

| Reshuffle rules | deck-out, fatigue | Reshuffle discard, or punish empty deck. |

| Removal / answers | counterplay | Every threat needs an answer in the pool. |

Patterns

1. Zones + draw with automatic reshuffle

# Pseudocode. A card is in exactly one zone at a time; moving = remove here, add there.
def draw(n):
    for _ in range(n):
        if not deck:
            if not discard:          # truly empty: deck-out (lose, or take fatigue)
                on_deck_out(); return
            deck.extend(discard)     # reshuffle discard into deck
            discard.clear()
            shuffle(deck, rng)       # use a seeded RNG (see save-systems for replays)
        hand.append(deck.pop())

2. Card as data + effect resolution

# Pseudocode. Effects are a data list interpreted by the engine — not bespoke code per card.
card = {
    "id": "fireball", "cost": 3, "type": "spell",
    "effects": [ {"op": "damage", "amount": 6, "target": "chosen_enemy"} ],
}
def play(card, caster):
    if resources[caster] < card.cost: return False     # can't afford
    resources[caster] -= card.cost
    move(card, from_zone=hand, to_zone=play_or_discard(card))
    for fx in card.effects:
        resolve_effect(fx, caster)                      # one interpreter handles every card
    return True

3. Turn structure as a phase machine

# Pseudocode. Fixed phases keep timing windows (triggers, priority) unambiguous.
PHASES = ["untap", "draw", "main", "combat", "end"]
def take_turn(player):
    for phase in PHASES:
        enter_phase(player, phase)        # fire "on_phase" triggers here
        if phase == "draw":   draw(1)
        if phase == "main":   await player_plays_cards()
        if phase == "combat": resolve_combat()
        if phase == "end":    discard_to_hand_limit(player); clear_temporary_effects()

Pitfalls / failure modes

• A card existing in two zones at once → duplication/loss bugs. Enforce "exactly one zone";

move = remove-then-add, and assert no card appears twice.

• Forgetting to reshuffle → draws silently fail or crash on empty deck. Reshuffle discard,

or define deck-out/fatigue explicitly (Pattern 1).

• One function per card → unmaintainable and untestable. Make effects data interpreted

by a small set of operations (Pattern 2).

• Ambiguous effect order / simultaneous triggers → nondeterministic outcomes. Resolve in a

defined order (a queue or stack); document LIFO vs. FIFO (refs).

• Unseeded shuffle in a game that needs replays/undo → can't reproduce. Use a seeded RNG.
• No hand limit / no answers → degenerate hoarding or unbeatable threats. Add a hand cap and

ensure removal exists for every threat archetype.

• Targeting state leaks → a cancelled play leaves the board mid-targeting. Make play

atomic: validate cost + targets first, then commit.

Composition (build it from these skills)

• Card content: godot-resources / unity-scriptableobjects — define each card as a data asset.
• UI: game-ui-ux for layout, scaling, and focus navigation; godot-ui-control for hand layout, drag/drop, zone counts, and targeting prompts.
• Persistence/replays: save-systems for collection, run state (roguelike deckbuilder), and seeded replays.
• Opponent AI: game-ai for an AI that evaluates playable cards and picks targets.
• Animation/feedback: the engine animation skill for card movement; audio-design for cues.
• Scripting: godot-gdscript / unity-csharp-scripting for the effect interpreter.

References

• For the effect queue/stack, keywords/triggers, targeting, deckbuilder vs. constructed

archetypes, and shuffle fairness, read references/effect-resolution.md.