スキル一覧に戻る

grade-tcg-card

pjt222
更新日 Yesterday
3 閲覧
17
2
17
GitHubで表示
その他general

について

このClaudeスキルは、トレーディングカード(ポケモン、MTG、Flesh and Blood、Kayouなど)をPSA、BGS、またはCGCの基準で鑑定します。センタリング、表面状態、エッジ、コーナーを体系的に評価し、信頼区間付きの最終グレードを決定します。鑑定提出前の事前審査、状態に関する議論の解決、価格幅の推定などにご利用いただけます。

クイックインストール

Claude Code

推奨
メイン
npx skills add pjt222/agent-almanac -a claude-code
プラグインコマンド代替
/plugin add https://github.com/pjt222/agent-almanac
Git クローン代替
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/grade-tcg-card

このコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします

ドキュメント

Grade TCG Card

Grade card per PSA/BGS/CGC. Observation-first (from meditate) → no grade anchoring.

Use When

  • Pre-submission evaluation
  • Pre-screen collection → high-grade candidates
  • Settle buyer/seller disputes
  • Learn consistent grading
  • Estimate grade-value spread

In

  • Required: card ID (set, #, name, variant)
  • Required: images or physical desc (front + back)
  • Required: standard (PSA 1-10, BGS 1-10 w/ subgrades, CGC 1-10)
  • Optional: known market value at grades
  • Optional: game (Pokemon, MTG, FaB, Kayou)

Do

Step 1: Clear bias → observe w/o prejudgment

From meditate Step 2-3: observe w/o anchor to expected grade/value.

  1. Set aside market value
  2. DO NOT look up sales/population reports pre-grade
  3. Known valuable → explicit: "Worth $X PSA 10. Set aside."
  4. Examine as physical object
  5. Gut impression noted but NOT anchored
  6. Premature grade thoughts → "anchoring" → return to observation

→ Neutral start. Grade anchoring = #1 source of inconsistency.

If err: bias sticky (high-value wants 10) → write bias explicit. Externalize → reduce influence. Proceed only as physical object.

Step 2: Centering

  1. Measure border all 4 sides front:
    • L vs R (horizontal)
    • T vs B (vertical)
    • Ratio: 55/45 L-R, 60/40 T-B
  2. Repeat back
  3. Apply thresholds:
PSA Centering Thresholds:
+-------+-------------------+-------------------+
| Grade | Front (max)       | Back (max)        |
+-------+-------------------+-------------------+
| 10    | 55/45 or better   | 75/25 or better   |
| 9     | 60/40 or better   | 90/10 or better   |
| 8     | 65/35 or better   | 90/10 or better   |
| 7     | 70/30 or better   | 90/10 or better   |
+-------+-------------------+-------------------+

BGS Centering Subgrade:
+------+-------------------+-------------------+
| Sub  | Front (max)       | Back (max)        |
+------+-------------------+-------------------+
| 10   | 50/50 perfect     | 50/50 perfect     |
| 9.5  | 55/45 or better   | 60/40 or better   |
| 9    | 60/40 or better   | 65/35 or better   |
| 8.5  | 65/35 or better   | 70/30 or better   |
+------+-------------------+-------------------+
  1. Record ratio per axis + subgrade.

→ Numeric ratios both faces + grade/subgrade. Most objective measurement.

If err: borders too narrow (full-art, borderless) → "centering N/A — borderless", skip Step 3. Services differ for borderless.

Step 3: Surface

  1. Front under good light:
    • Print defects: ink spots, missing ink, print lines, color inconsistency
    • Scratches: direct + angled light
    • Whitening: haze/clouding
    • Indentations: raking light
    • Stain/discoloration: yellowing, water marks, chemical
  2. Back same criteria
  3. Factory vs handling:
    • Factory: print lines, miscut, crimping → may be less penalized
    • Handling: scratches, dents, stains → always penalized
  4. Rate:
    • Pristine (10): flawless under mag
    • Near-pristine (9-9.5): minor only under mag
    • Excellent (8-8.5): minor naked eye
    • Good (6-7): moderate, multiple
    • Fair+ (1-5): significant

→ Defect inventory located + described + severity. Factory vs handling distinguished.

If err: low-res images → provide range not point grade. Recommend physical.

Step 4: Edge + corner

  1. 4 edges:
    • Whitening: spots/lines along colored (most common)
    • Chipping: edge layer missing
    • Roughness: uneven / micro-tears
    • Foil separation: holofoil delamination
  2. 4 corners:
    • Sharpness: tip crisp
    • Rounding: worn curve (slight/moderate/heavy)
    • Splitting: layer separation (dings)
    • Bending: turned/creased
  3. Rate same scale
  4. Note worst edge/corner

→ Per-edge + per-corner. Worst limits overall grade.

If err: in sleeve/toploader obscures → note which areas not fully assessed.

Step 5: Final grade

  1. PSA (1-10):
    • Weakest sub limits final
    • Perfect surface + 65/35 centering → cap PSA 8
    • "Lowest limits" + adjust up if exceptional
  2. BGS (4 subgrades → overall):
    • Centering, Edges, Corners, Surface each 1-10 in 0.5
    • Overall = weighted avg, lowest subgrade limits
    • BGS 10 Pristine → all 4 = 10
    • BGS 9.5 Gem Mint → avg 9.5+ no sub <9
  3. CGC (similar to PSA w/ subgrades on label):
    • Centering, Surface, Edges, Corners
    • Overall = proprietary weighting
  4. Confidence:
    • "PSA 8 (confident)" — clear, unlikely higher/lower
    • "PSA 8-9 (borderline)" — either way
    • "PSA 7-8 (uncertain)" — limited data

→ Final grade + confidence. BGS all 4 reported. Evidence from Steps 2-4.

If err: inconclusive (scratch vs dirt) → range + recommend pro grading. Never confident w/ insufficient data.

Check

  • Bias check pre-grading (no anchoring)
  • Centering both faces w/ ratios
  • Surface examined (scratches, print, stain, indent)
  • All 4 edges + corners individually
  • Factory vs handling distinguished
  • Grade evidence-backed from each sub
  • Confidence stated
  • Standard applied correctly

Traps

  • Grade anchoring: value knowledge → bias toward hoped-for grade. Physical first.
  • Ignore back: back surface + centering count. Many over-focus front.
  • Factory vs handling confusion: both affect grade differently.
  • Holofoil over-grade: hides scratches until angled. Multiple light angles.
  • Centering illusion: art placement can mislead. Measure borders not art.

  • build-tcg-deck — condition affects tournament legality
  • manage-tcg-collection — grade-based valuation
  • meditate — source of observation-without-prejudgment

GitHub リポジトリ

pjt222/agent-almanac
パス: i18n/caveman-ultra/skills/grade-tcg-card
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

関連スキル

llamaguard

その他

LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。

スキルを見る

cost-optimization

その他

このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。

スキルを見る

quantizing-models-bitsandbytes

その他

このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。

スキルを見る

dispatching-parallel-agents

その他

このClaudeスキルは、複数のエージェントを配備し、3つ以上の独立した問題を並行して調査・修正します。共有状態や依存関係がなく解決可能な、無関係な障害が発生するシナリオ向けに設計されています。中核となる機能は並列問題解決であり、効率を最大化するために独立した問題領域ごとに1つのエージェントを割り当てます。

スキルを見る