name	ai-slay-the-spire-2
description	Strategy workflow for piloting Slay the Spire 2 runs with an AI agent. Use when deciding combat plays, card rewards, route choices, shops, rest sites, relics, potions, or post-run tuning for Slay the Spire 2, especially Ironclad-style warrior runs.

AI Slay the Spire 2

Use this skill as a compact decision layer. Load detailed references only when the current game state needs them.

Core Loop

Read the current game state: act, floor, HP, deck, relics, potions, map, enemies, rewards, and known boss.
Evaluate deck readiness with references/deck_strength_evaluation.md.
Identify weakness_tags: damage, block, scaling, draw, aoe, control, elite_survival, boss_solution.
Evaluate the next 3-5 floors as a run segment: route risk, rest/shop access, boss needs, and expected HP loss.
Choose the action that solves the nearest real threat without damaging long-term consistency.
Record short reasoning when a decision is non-obvious so later runs can tune parameters.

Reference Loading

For route, shop, rest-site, potion, boss, and high-level strategy, read references/broken_star_sts_strategy.md.
For card rewards, read references/card_selection_scoring.md after deck readiness is known.
For deck strength and route readiness, read references/deck_strength_evaluation.md.
For card and relic tags, read data/cards_warrior.yaml and data/relic_modifiers.yaml.
For unvalidated local learnings, inspect proposals/ only after official rules are insufficient.
For regression examples, read tests/decision_fixtures.yaml.

Decision Priorities

Combat:

if a lethal-safe kill exists:
    kill
elif incoming damage is lethal or causes route failure:
    block, weak, potion, or remove the attacking threat
elif enemy has a special layer/damage-cap mechanic:
    spend cheap or multi-hit attacks to remove layers
elif a scaling play is safe before the next attack cycle:
    play scaling
else:
    balance damage, block, setup, and potion conservation

Card rewards:

evaluate deck readiness first
score each card against weakness_tags, boss needs, relic synergies, deck pollution, and role saturation
score skip as a real fourth option before comparing cards
if attack_count is already sufficient and weakness_tags lacks damage/aoe:
    penalize additional attacks unless they are premium, synergistic, or solve a known boss/elite
if deck lacks reliable block/control or HP loss is high:
    raise block, weak, draw, and consistency above more frontload
pick only if best_card_score >= skip_score + pick_margin and the card improves the next run segment
otherwise skip

Route:

prefer the route that converts current resources into survival plus scaling
avoid elite paths when route_readiness_state is risky or danger unless potion/rest support is clear
value shops when gold can become removal, potions, or a concrete shortboard answer
protect enough HP to reach the act boss with a functional deck, not just to win the next fight

Guardrails

Do not force an imagined archetype before the deck has survived the current act pressure.
Do not treat "more damage" as the default answer after the deck already has enough attacks for the next route segment.
Do not take medium attacks over skip when the real weaknesses are block, control, draw, consistency, or HP preservation.
Do not pick from card rewards by choosing the least bad card; skip is preferred when no card solves a current or future problem.
Do not let deck size grow without a compensating reason such as draw, energy, exhaust, removal, or a high-impact solution.
Do not spend HP for short-term output when the next floors require survival, rest access is distant, or the boss plan is incomplete.
Do not leave a shop automatically when gold can buy removal, a relevant potion, or a shortboard answer.
Do not save potions through lethal or large preventable HP loss.
Do not overvalue medium transition cards once the deck already answers the next route segment.
Treat proposal files as hypotheses, not official rules, until multiple runs support them.

Output Format

Use compact decision notes:

Decision: pick/remove/buy/rest/smith/play/route
Need: damage/block/scaling/draw/aoe/control/boss/none
Readiness: strong/acceptable/risky/danger
Reason: one sentence tied to the immediate threat and long-term cost.

AI Slay the Spire 2

Use this skill as a compact decision layer. Load detailed references only when the current game state needs them.

Core Loop

Read the current game state: act, floor, HP, deck, relics, potions, map, enemies, rewards, and known boss.

Evaluate deck readiness with references/deck_strength_evaluation.md.

Identify weakness_tags: damage, block, scaling, draw, aoe, control, elite_survival, boss_solution.

Evaluate the next 3-5 floors as a run segment: route risk, rest/shop access, boss needs, and expected HP loss.

Choose the action that solves the nearest real threat without damaging long-term consistency.

Record short reasoning when a decision is non-obvious so later runs can tune parameters.

Reference Loading

For route, shop, rest-site, potion, boss, and high-level strategy, read references/broken_star_sts_strategy.md.

For card rewards, read references/card_selection_scoring.md after deck readiness is known.

For deck strength and route readiness, read references/deck_strength_evaluation.md.

For card and relic tags, read data/cards_warrior.yaml and data/relic_modifiers.yaml.

For unvalidated local learnings, inspect proposals/ only after official rules are insufficient.

For regression examples, read tests/decision_fixtures.yaml.

Decision Priorities

Combat:

if a lethal-safe kill exists: kill elif incoming damage is lethal or causes route failure: block, weak, potion, or remove the attacking threat elif enemy has a special layer/damage-cap mechanic: spend cheap or multi-hit attacks to remove layers elif a scaling play is safe before the next attack cycle: play scaling else: balance damage, block, setup, and potion conservation

Card rewards:

evaluate deck readiness first score each card against weakness_tags, boss needs, relic synergies, deck pollution, and role saturation score skip as a real fourth option before comparing cards if attack_count is already sufficient and weakness_tags lacks damage/aoe: penalize additional attacks unless they are premium, synergistic, or solve a known boss/elite if deck lacks reliable block/control or HP loss is high: raise block, weak, draw, and consistency above more frontload pick only if best_card_score >= skip_score + pick_margin and the card improves the next run segment otherwise skip

Route:

prefer the route that converts current resources into survival plus scaling avoid elite paths when route_readiness_state is risky or danger unless potion/rest support is clear value shops when gold can become removal, potions, or a concrete shortboard answer protect enough HP to reach the act boss with a functional deck, not just to win the next fight

Guardrails

Do not force an imagined archetype before the deck has survived the current act pressure.

Do not treat "more damage" as the default answer after the deck already has enough attacks for the next route segment.

Do not take medium attacks over skip when the real weaknesses are block, control, draw, consistency, or HP preservation.

Do not pick from card rewards by choosing the least bad card; skip is preferred when no card solves a current or future problem.

Do not let deck size grow without a compensating reason such as draw, energy, exhaust, removal, or a high-impact solution.

Do not spend HP for short-term output when the next floors require survival, rest access is distant, or the boss plan is incomplete.

Do not leave a shop automatically when gold can buy removal, a relevant potion, or a shortboard answer.

Do not save potions through lethal or large preventable HP loss.

Do not overvalue medium transition cards once the deck already answers the next route segment.

Treat proposal files as hypotheses, not official rules, until multiple runs support them.

Output Format

Use compact decision notes:

Decision: pick/remove/buy/rest/smith/play/route Need: damage/block/scaling/draw/aoe/control/boss/none Readiness: strong/acceptable/risky/danger Reason: one sentence tied to the immediate threat and long-term cost.

ai-slay-the-spire-2