STS2 Expert — Agente Avaliador de Cartas e Relíquias
ing-gom/sts2-expert
Avaliador expert de StS2 com catalogação e grades automáticas
Ver no GitHubAgente especialista que cataloga e avalia cartas, relíquias e poções do Slay the Spire 2 usando Claude Code. Fornece grades S–D baseadas em eficiência, escalabilidade e sinergia de deck, com visualizador HTML interativo que permite filtrar e consultar a base de dados completa do jogo.
Instalação
npx convez-ai ing-gom/sts2-expert1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliar eficiência e potencial de cartas e relíquias específicas
- Pesquisar e filtrar o catálogo completo de cards por raridade e classe
- Verificar sinergia e valor esperado de construções de deck
- Consultar critérios de avaliação baseados em meta-jogo e mecânicas
Veredito

Ferramenta especializada e bem-estruturada para fãs sérios de StS2 que combina avaliação sistemática com acesso aos dados brutos do jogo. Excelente documentação técnica, mas nicho de uso (específico do jogo e da comunidade de StS2).
README
A tradução para o português deste README ainda está na fila.
STS2 Expert — Card/Relic/Potion Evaluation Agent & Catalog
English · 한국어
A public bundle containing a Claude Code expert agent that catalogs and evaluates the cards, relics, and potions of Slay the Spire 2 (Early Access), the data snapshots the agent relies on, and a browser-based HTML catalog viewer.
📊 Catalog viewer: https://ing-gom.github.io/sts2-expert/ — Every card, relic, and potion shows an S–D grade badge, with grade sorting, grade filtering, and a one-line rationale.
Layout
.claude/agents/sts2-expert.md # Evaluation expert agent (read-only evaluator)
.claude/skills/
tier-judge/ # Per-cost efficiency table + grading rubric
card-expected-value/ # 5-turn EV (Expected Value) measurement basis
card-effect-verify/ # Close-reading verification procedure for descriptions
eval_criteria.md # Evaluation criteria (efficiency, scaling, new mechanics, character meta)
data/ # Game data snapshots (from headless dumps)
index.html # Catalog viewer (GitHub Pages)
Using the agent
With Claude Code installed:
git clone https://github.com/ing-gom/sts2-expert
cd sts2-expert
claude
Opening Claude Code inside the repo auto-registers the sts2-expert agent. Examples:
"Evaluate BLOOD_POTION with the sts2-expert agent"
"Catalog and grade all Silent rare cards"
"How good is the Akabeko relic?"
Evaluation principles
The agent is a read-only evaluator — it organizes information from the data and returns reports with S–D grades and rationale; it never modifies files.
- Standalone judgment: Grades are derived solely from a card's factual effects and the
rubric in
eval_criteria.md. External evaluation data such as community tier lists (tier/scorefields, external tier CSVs) has been removed from the bundle and is never consulted or cited at any stage. (The empirical causal data incard_empirical_lift.jsonis a measurement, not an external opinion, so it may be cited.) - Static vs. contextual separation: Static grades (ceiling) and contextual grades (deck/timing) are always evaluated separately.
- Criteria upkeep:
eval_criteria.mdholds the efficiency tables, scaling, deck-thinning, new mechanics (Corrosion/Penetration/Durability), conditional penalties, and character-meta criteria — it captures evaluation methodology and meta, not tier values.
The grades published in the viewer are this agent's standalone evaluation results
(data/grades.json).
Data
| File | Contents |
|---|---|
data/cards_catalog.json | Card master catalog (1,117 incl. upgrades, game_version v0.103.2 — tier/score fields removed) |
data/grades.json | Agent standalone S–D grades (577 cards · 309 relics · 64 potions; grade + context + one-line rationale; optional per-upgrade grade) |
data/card_expected_value.json | Card EV — 5-turn baseline damage equivalent |
data/card_axis_overrides.json | Explicit axis tags (Producer/Consumer/Amplifier/Trigger) |
data/card_empirical_lift.json | Full-run measured causal card lift (not an external opinion) |
data/role_needs.json | Cross-axis synergy direction/weights |
data/build_completion.json | Per-character/archetype keystone definitions |
data/effect_value_table.json | Value-conversion table by effect type |
data/axis_labels_ko.json | Korean axis labels |
data/relics_locale.json | 309 relics, ko/en/zh names + descriptions (game_version v0.103.3) |
data/relic_canonical_vars.json | Relic placeholder real values (constants extracted from the game DLL) |
data/relic_axis_overrides.json | Relic → card axis synergy mapping |
data/potions.json | 64 potions — ko/en/zh names/descriptions + Rarity/Usage/Target + placeholder real values (headless dump) + curated classification (game_version v0.103.3) |
Dump versions differ by component — cards are the stable v0.103.2 snapshot; relics and potions are v0.103.3.
Running the viewer locally
index.html fetches data/*.json, so it cannot be opened via file://:
python -m http.server 8000
# http://localhost:8000
License / notice
An unofficial fan project. Rights to the game data belong to Mega Crit. Given Early Access, data snapshots may differ from the latest game balance. The tooling code (agent/skills/viewer) is under the MIT license.