agentleFS
phase-rsphase-rs/phase/.claude/skills/add-card-data-pipeline/SKILL.md

add-card-data-pipeline

Skill331 starsChanged 4 months agoApache-2.0View on GitHub

Use when modifying the card data pipeline — adding new fields to card exports, changing Oracle parser output shape, updating card database loading, modifying the coverage report, adding synthesis functions, or debugging why a card's data looks wrong at runtime.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before you start: Run the pipeline end-to-end once to see what it produces: cargo run --release --bin oracle-gen -- data --stats > …

    line 15
  2. Distinctive:

    When a new feature needs data that isn't currently exported:

    line 204
  3. Very specific:

    Add the field. Use #[serde(default)] for backward compatibility — existing card-data.json files must still deserialize.

    line 207
  4. Distinctive:

    Add the optional field to the card type definition.

    line 217
  5. Distinctive:

    When a keyword or ability implies game mechanics that Oracle text doesn't make explicit:

    line 221
  6. Very specific:

    Add the call in synthesize_all() alongside existing synthesis calls.

    line 227
  7. Very specific:

    When a parser change affects the structure of ParsedAbilities:

    line 233
  8. Distinctive:

    After completing work using this skill:

    line 287
  9. Distinctive:

    Update the synthesis function list: if you added a new one

    line 290
  10. Distinctive:

    Update the CardFace fields: if the struct changed

    line 291

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Card Data Pipeline

Hard rules — all pipeline work must respect these (see CLAUDE.md § Design Principles):

  1. CR-correctness is non-negotiable. The pipeline is the bridge between Oracle text and engine types. Parser output must faithfully represent the Comprehensive Rules semantics of each card. If a synthesis function or parser pattern produces an incorrect typed representation, the engine will enforce wrong rules. Verify every new pattern against the relevant CR section and annotate the code.
  2. Build for the class, not the card. Every parser pattern, synthesis function, and export field must handle a category of cards. A synthesis function that works for one card but mishandles the next card with the same keyword pattern is a defect. Test with multiple representative cards from the class.
  3. Test the building block. Every new parser pattern needs a unit test. Every new synthesis function needs a test verifying its output. Run cargo coverage after changes to verify coverage improvements. Run cargo insta review if snapshot tests are affected.

The card data pipeline converts raw MTGJSON card data into the typed card-data.json consumed by the WASM frontend and multiplayer server. Understanding this pipeline is essential when adding new ability types, fields, or synthesis steps — a change to the export format must propagate through all loading paths.

Before you start: Run the pipeline end-to-end once to see what it produces: cargo run --release --bin oracle-gen -- data --stats > /tmp/test-export.json 2>/tmp/stats.txt && head -1 /tmp/stats.txt

CR Verification Rule: Every CR number in annotations MUST be verified by grepping docs/MagicCompRules.txt before writing. Do NOT rely on memory — 701.x and 702.x numbers are arbitrary sequential assignments that LLMs consistently hallucinate. Run grep -n "^701.21" docs/MagicCompRules.txt (etc.) for every number. If you cannot find it, do not write the annotation.


Pipeline Flow

MTGJSON (AtomicCards.json, ~30k cards)
    ↓ download: scripts/gen-card-data.sh
data/mtgjson/AtomicCards.json
    ↓ parse: crates/engine/src/database/mtgjson.rs
HashMap<String, Vec<AtomicCard>>
    ↓ process: crates/engine/src/database/synthesis.rs (build_oracle_face)
    │   For each card face:
    │   ├─ build_card_type() → CardType
    │   ├─ parse_oracle_text() → ParsedAbilities { abilities, triggers, statics, replacements }
    │   └─ synthesize_all() — runs all synthesis functions (see function for current list)
    ↓ export: crates/engine/src/bin/oracle_gen.rs
client/public/card-data.json (~49 MB in the current export)
    ↓ consume (three loading paths):
    ├─ WASM: CardDatabase::from_json_str()  [browser, via engine-wasm]
    ├─ Server: CardDatabase::from_export()   [phase-server, multiplayer]
    └─ CI: CardDatabase::from_export()       [coverage-report binary]

Quick Commands

Regenerate card data:

cargo run --release --bin oracle-gen -- data --stats > client/public/card-data.json

Look up a specific card's exported data:

jq '.["card name in lowercase"]' client/public/card-data.json
# Example:
jq '.["lightning bolt"]' client/public/card-data.json

Check if a card has unimplemented abilities:

jq '.["card name"] | {abilities: [.abilities[]? | select(.effect.type == "Unimplemented")], triggers: [.triggers[]? | select(.mode == "Unknown")]}' client/public/card-data.json

Run coverage report:

cargo coverage                                    # Standard-legal cards
cargo run --bin coverage-report -- data --all     # All cards
cargo run --bin coverage-report -- data --ci      # CI mode (exits 1 on gaps)

Key Files

Data Source — MTGJSON

crates/engine/src/database/mtgjson.rs — Deserialization types for MTGJSON format.

pub struct AtomicCardsFile {
    pub data: HashMap<String, Vec<AtomicCard>>,  // name → faces
}

pub struct AtomicCard {
    pub name: String,
    pub mana_cost: Option<String>,        // "{2}{R}" format
    pub colors: Option<Vec<String>>,       // ["R", "G"]
    pub types: Vec<String>,               // ["Creature"]
    pub subtypes: Option<Vec<String>>,     // ["Elf", "Warrior"]
    pub supertypes: Option<Vec<String>>,   // ["Legendary"]
    pub power: Option<String>,            // "3" or "*"
    pub toughness: Option<String>,
    pub text: Option<String>,             // Oracle text
    pub keywords: Option<Vec<String>>,     // ["Flying", "Ward"]
    pub layout: Option<String>,           // "normal", "transform", "split", etc.
    pub legalities: Option<HashMap<String, String>>,
    pub identifiers: Option<Identifiers>,  // includes scryfall_oracle_id
}

Synthesis — crates/engine/src/database/synthesis.rs

build_oracle_face(mtgjson, oracle_id) → CardFace — Converts one MTGJSON face into a typed CardFace:

  1. build_card_type() — maps type strings to CardType { supertypes, core_types, subtypes }
  2. parse_oracle_text() — runs Oracle parser → ParsedAbilities
  3. synthesize_all() — runs all keyword-implied synthesis (see synthesize_all() for the current registry). Each synthesize_* function takes &mut CardFace and adds abilities/triggers/statics that the keyword implies but Oracle text doesn't make explicit.

Oracle Loader — crates/engine/src/database/oracle_loader.rs

Handles MTGJSON-specific loading concerns: card layout detection, face splitting for DFCs/split cards, and legality normalization. Delegates to synthesis.rs for the per-face build pipeline.

Export Binary — crates/engine/src/bin/oracle_gen.rs

Orchestrates the full export:

  1. Load MTGJSON via load_atomic_cards()
  2. For each card: build_oracle_face() per face
  3. Map layout to CardLayout enum (Single, Split, Transform, etc.)
  4. Normalize legalities
  5. Flatten to HashMap<String, CardExportEntry> (key = lowercase face name, value = flattened CardFace + legalities)
  6. Serialize to JSON → stdout

Card Database — crates/engine/src/database/card_db.rs

Three loading methods:

MethodUsed ByWhat It Loads
from_mtgjson(path)oracle_gen, testsFull parse — runs Oracle parser on raw MTGJSON
from_export(path)phase-server, coverage-reportPre-processed card-data.json with flattened face entries + legalities
from_json_str(json)engine-wasm (browser), deck validation testsSame as from_export but takes string input

Internal structure:

pub struct CardDatabase {
    pub cards: HashMap<String, CardRules>,     // Populated by from_mtgjson
    pub face_index: HashMap<String, CardFace>, // Populated by all loading paths
    pub legalities: HashMap<String, CardLegalities>,
    pub errors: Vec<(PathBuf, String)>,
}

Card Face — crates/phase-ast/src/types/card.rs

The serialized unit in card-data.json:

pub struct CardFace {
    pub name: String,
    pub mana_cost: ManaCost,
    pub card_type: CardType,
    pub power: Option<PtValue>,
    pub toughness: Option<PtValue>,
    pub loyalty: Option<String>,
    pub defense: Option<String>,
    pub oracle_text: Option<String>,
    pub non_ability_text: Option<String>,
    pub flavor_name: Option<String>,
    pub keywords: Vec<Keyword>,
    pub abilities: Vec<AbilityDefinition>,
    pub triggers: Vec<TriggerDefinition>,
    pub static_abilities: Vec<StaticDefinition>,
    pub replacements: Vec<ReplacementDefinition>,
    pub color_override: Option<Vec<ManaColor>>,
    pub scryfall_oracle_id: Option<String>,
    pub modal: Option<ModalChoice>,
    pub additional_cost: Option<AdditionalCost>,
    pub casting_restrictions: Vec<CastingRestriction>,
    pub casting_options: Vec<SpellCastingOption>,
}

Coverage Report — crates/engine/src/bin/coverage_report.rs

Checks each card face for:

  • Keyword::Unknown(_) → unsupported keyword
  • Effect::Unimplemented { .. } → unsupported effect
  • TriggerMode::Unknown(_) → unsupported trigger
  • Unrecognized StaticMode → unsupported static

Outputs JSON summary: { total_cards, supported_cards, coverage_pct, cards: [...], missing_handler_frequency: [...] }

Important current behavior:

  • coverage-report loads card-data.json via CardDatabase::from_export(), not raw MTGJSON
  • Without --all, it filters to standard-cards.txt
  • It strips known-benign MTGJSON keyword mismatches (bare parameterized keywords like Keyword:Ward and action-keyword noise like Keyword:Scry) before computing final manifest coverage
  • --write-warning-patterns <path> writes clustered parser warnings ranked by likely shared fix
  • --warning-pattern '<pattern>' emits a compact JSON drilldown for one exact parser warning pattern
  • --warning-detector <detector> emits a compact JSON drilldown for a broader swallowed-clause detector family
  • --warning-full includes the full parse_details tree and exported CardFace JSON in drilldown output

CI mode (--ci): exits 1 if any Standard-legal cards have gaps.

Coverage Module — crates/engine/src/game/coverage.rs

pub fn unimplemented_mechanics(obj: &GameObject) -> Vec<String>

Checks a game object at runtime for unsupported mechanics. Used by the frontend to show amber "!" badges on cards with partial support.


Checklist — Modifying the Pipeline

Adding a New Field to Card Export

When a new feature needs data that isn't currently exported:

  • crates/phase-ast/src/types/card.rs — CardFace struct Add the field. Use #[serde(default)] for backward compatibility — existing card-data.json files must still deserialize.
  • crates/engine/src/database/synthesis.rs — build_oracle_face() Populate the new field during card processing.
  • crates/engine/src/bin/oracle_gen.rs — Usually no changes needed (serializes CardFace automatically via serde).
  • Regenerate card-data.json: cargo run --release --bin oracle-gen -- data --stats > client/public/card-data.json
  • client/src/adapter/types.ts — TypeScript types (if frontend needs the field) Add the optional field to the card type definition.

Adding a New Synthesis Function

When a keyword or ability implies game mechanics that Oracle text doesn't make explicit:

  • crates/engine/src/database/synthesis.rs — new synthesize_*() function Pattern: takes &mut CardFace, checks for the triggering condition (keyword, type, etc.), adds abilities/triggers/statics to the face.
  • crates/engine/src/database/synthesis.rs — call from synthesize_all() Add the call in synthesize_all() alongside existing synthesis calls. This is automatically invoked by build_oracle_face().
  • Test: Add a test in synthesis.rs verifying the synthesis produces the expected output.

Modifying the Oracle Parser Output Shape

When a parser change affects the structure of ParsedAbilities:

  • Update all three loading paths — from_mtgjson, from_export, and from_json_str all need to understand the serialized CardFace shape. If you add/rename fields, verify both raw-MTGJSON loading and flattened export loading.
  • Audit serialized consumers — If the parser output shape changes an enum or serialized ability field, verify WASM JSON loading, server from_export, coverage-report loading, and any AI/community scenario fixtures that store the changed enum. Add serde defaults, fixture migrations, or compatibility only when deliberately needed.
  • Regenerate card-data.json — Always regenerate after parser changes.
  • Update coverage report — If the parser now recognizes previously-unimplemented patterns, coverage numbers will change. Run the report to verify improvements, and update any benign-keyword filtering in coverage_report.rs if the mismatch profile changed.
  • Coverage honesty gate — Accepting full Oracle text while dropping a rules-bearing rider, continuation, restriction, granted ability, replacement, or sentence tail is a hard failure unless an Effect::unimplemented, TriggerMode::Unknown, equivalent strict-failure marker, or unchanged unsupported coverage keeps the card red. Coverage improvements must list shipped semantics separately from residual unsupported riders.
  • Update snapshot tests — crates/engine/tests/oracle_parser.rs has insta snapshots that must be updated: cargo insta review.

CI Coverage Gate

The coverage report checks card support across format-based legality data. There is no separate manifest file — coverage is computed from the cards' legality fields in card-data.json.


WASM Loading Path

Understanding how card data reaches the browser:

  1. Build time: oracle-gen produces client/public/card-data.json
  2. Runtime: Frontend fetches /card-data.json via HTTP
  3. WASM init: load_card_database(json_str) in crates/engine-wasm/src/lib.rs
  4. Parse: CardDatabase::from_json_str() → deserializes flattened export entries into face_index + normalized legalities
  5. Storage: Thread-local CARD_DB: RefCell<Option<CardDatabase>>
  6. Usage: initialize_game() resolves deck card names via face_index lookup

Important: The WASM bridge receives the JSON as a string from JavaScript, not as a file path. Any changes to the JSON structure must round-trip through serde correctly.


Common Mistakes

MistakeConsequenceFix
Missing #[serde(default)] on new CardFace fieldsExisting card-data.json fails to deserializeAlways default new optional fields
Changing field types without regenerating card-data.jsonRuntime deserialization panicAlways regenerate after type changes
Adding synthesis but not calling from build_oracle_face()Synthesis exists but never runs — abilities missing from exportAdd the call
Not updating TypeScript typesFrontend can't access new fields, or gets wrong typesUpdate adapter/types.ts
Modifying CardFace but only testing from_mtgjson pathfrom_export and from_json_str may fail differentlyTest all three paths
Parser accepts a full sentence but drops a semantic riderCoverage claims support while the engine enforces incomplete rulesPreserve an explicit Unimplemented/Unknown/strict-failure marker or implement the whole rules-bearing clause
Parser enum shape changes but AI/community fixtures are staleSerialized scenario loading fails even though parser/card-data checks passMigrate fixtures or provide an intentional compatibility path
Coverage looks wrong after parser improvementManifest filtering or benign-keyword stripping hides real resultCheck coverage_report.rs filtering logic
Forgetting to regenerate after parser changesCard-data.json contains stale parsed dataRun oracle-gen after any parser modification
Parser produces new support but coverage still looks wrongManifest filtering or benign MTGJSON keyword mismatches hide the real resultCheck coverage_report.rs filtering before assuming the parser regressed

Self-Maintenance

After completing work using this skill:

  1. Verify references with the check below
  2. Update the synthesis function list if you added a new one
  3. Update the CardFace fields if the struct changed

Verification

rg -q "fn build_oracle_face" crates/engine/src/database/synthesis.rs && \
rg -q "fn synthesize_basic_land_mana" crates/engine/src/database/synthesis.rs && \
rg -q "fn synthesize_equip" crates/engine/src/database/synthesis.rs && \
rg -q "fn synthesize_changeling_cda" crates/engine/src/database/synthesis.rs && \
rg -q "fn synthesize_all" crates/engine/src/database/synthesis.rs && \
rg -q "fn from_export" crates/engine/src/database/card_db.rs && \
rg -q "fn from_json_str" crates/engine/src/database/card_db.rs && \
rg -q "fn from_mtgjson" crates/engine/src/database/card_db.rs && \
rg -q "struct CardFace" crates/phase-ast/src/types/card.rs && \
rg -q "fn parse_oracle_text" crates/phase-parser/src/parser/oracle.rs && \
rg -q "fn unimplemented_mechanics" crates/engine/src/game/coverage.rs && \
test -f crates/engine/src/bin/oracle_gen.rs && \
test -f crates/engine/src/bin/coverage_report.rs && \
test -f crates/engine/src/database/synthesis.rs && \
echo "✓ add-card-data-pipeline skill references valid" || \
echo "✗ STALE — update skill references"

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.