| name | lorcana-cards |
| description | Single owner for everything related to a specific Disney Lorcana card in `packages/lorcana/lorcana-cards` — authoring, debugging, triaging, generating, and any bounded engine or types extension required to make the printed text work. Use whenever the unit of work is "this card". |
Lorcana Cards
This skill owns the full lifecycle of a single card: from printed text to passing test, including bounded extensions to lorcana-engine and lorcana-types when the card requires them. There is no separate "engine-extend" skill for card-driven gaps. If a card needs the engine to do something new, that work lives here.
Use this skill when the unit of work is a specific card. For cross-cutting engine work (new variant unsupported by any card), follow the engine-package conventions in packages/lorcana/lorcana-engine/AGENTS.md directly.
Required Memory Step
- Read
memory/schema.md. It is the canonical schema shared across all Lorcana skills.
- Read the Guardrails and Promoted Rules sections of
memory/bank.md.
- Apply guardrails before editing. After substantive work, follow the schema's Memory Update Protocol.
Scope and Boundaries
Owned by this skill:
- Authoring the card definition file (
{NUMBER}-{slug}.ts).
- Authoring or activating the test file (
{NUMBER}-{slug}.test.ts).
- Bounded engine/types extensions when a single card needs them.
- Classifying whether a failing card is an authoring, engine, stub, or test-surface problem.
- Maintenance of
missingImplementation / missingTests flags.
Not owned by this skill (escalate or use sibling skills):
- Generating brand-new test patterns from scratch with no card context →
lorcana-test-generation (provides harness reference; this skill orchestrates).
- Resolving rules ambiguity →
lorcana-rules. Consume its Mode B JSON; do not reason about CR text inline.
- Card identity lookup and similar-card retrieval →
lorcana-find-card. Always pass k explicitly.
- Cross-cutting engine refactors that span many cards or add a new variant family → engine-package owners; create a minimal repro and hand off.
The Loop (canonical workflow)
Resolve → Ground → Probe → Author → Test → Verify
↓ on failure
Triage → (back to one of the above)
1. Resolve
bun packages/lorcana/lorcana-cards/src/cards/similarity.ts \
--query "<card>" --limit 10
Pass k (here --limit) explicitly. Default is 10. If status: "ambiguous", ask the user to choose; never guess. Capture: definition path, test path, top-k similar cards filtered to active-test references only.
2. Ground
If the printed text touches a non-trivial rule (timing, replacement, alternate cost, multiplayer interaction), ask lorcana-rules for Mode B JSON. Use its behaviorConstraints and testImplications to drive authoring and assertions. Carry the citations into the test file as a top-of-file comment.
If the text is unambiguous (vanilla stats, plain keyword, simple "deal N damage"), skip grounding.
3. Probe
Run the support-probe before declaring an engine gap. It walks a proposed ability shape and reports which referenced variants are registered:
bun packages/lorcana/lorcana-engine/src/support-probe/cli.ts \
--ability '{"type":"triggered","trigger":{"event":"play","on":"SELF"},"effect":{...}}'
Or programmatically: import { probeSupport } from "@tcg/lorcana-engine/support-probe".
Exit code 0 = all referenced variants supported; 1 = at least one missing/unknown. Output is JSON with per-surface checks (effect, condition, target, trigger-event, trigger-subject) and the registry each was checked against.
For deeper investigation when the probe flags missing, consult the source:
- Effect resolvers:
packages/lorcana/lorcana-engine/src/runtime-moves/resolution/action-effects/
- Activated abilities:
packages/lorcana/lorcana-engine/src/runtime-moves/moves/abilities/activate-ability.ts
- Triggered:
packages/lorcana/lorcana-engine/src/triggered-abilities/index.ts
- Derived state and projection:
packages/lorcana/lorcana-engine/src/rules/derived-state.ts, packages/lorcana/lorcana-engine/src/projection
- Variant registries:
ACTION_EFFECT_RESOLVER_TYPES, CONDITION_VARIANT_TYPES, TARGET_VARIANT_TYPES
A claimed engine gap must reduce to one named registry entry or one named resolver. Anything vaguer is not yet a gap — keep probing. Note that the probe is structural (registered ≠ correct end-to-end runtime); it answers "is this discriminator handled at all", not "does this exact ability run cleanly".
4. Author
Reference: PATTERNS.md for DSL shapes; packages/lorcana/lorcana-cards/AGENTS.md for ability-type conventions. Do not duplicate those tables here.
Hard rules:
- Implement from printed text, not from a stale generated stub.
- If a bounded engine/types extension is needed, do it in the same task. Add or extend the per-variant test in
lorcana-engine (see engine AGENTS.md) so the extension is registry-tracked.
- Only leave
missingImplementation: true after Probe identifies an unbounded gap and you can name the exact unsupported registry entry.
- Enchanted/epic variants share
canonicalId and copy abilities verbatim from the base card.
5. Test
Use the harness reference in lorcana-test-generation. Hard rules:
- One active behavior test per printed clause.
- Tests asserting only
missingImplementation, missingTests, or empty abilities arrays do not count as coverage.
- Prefer semantic matchers (
toHaveZoneCounts, toHaveKeyword, toBeAtLocation) over raw state reads.
- Hidden-zone, under-card, and chooser flows require authoritative state assertions, not client snapshots.
6. Verify
bun test --cwd packages/lorcana/lorcana-cards "./src/cards/<SET>/<TYPE>/<NUMBER>-<slug>.test.ts"
bun run --cwd packages/lorcana/lorcana-cards check-types
Widen scope only when the change touched shared engine code:
- Engine variant tests for any registry entry you extended.
- Card-side neighbors that reuse the same resolver/helper.
Triage (on failure)
Classify the failure into exactly one bucket before changing code:
| Bucket | Signal | Next step |
|---|
AUTHORING | Wrong DSL shape, owner scope, duration, or missing-flag | Fix the card file; rerun |
ENGINE_GAP | Probe identifies a missing registry entry or resolver branch | Bounded extension in same task |
STUB_QUALITY | Generated file is structurally wrong (parser/generator output) | Fix the generator if blocker repeats; otherwise hand-author and flag in observations |
TEST_SURFACE | Empty file, legacy comments, keyword-smoke-only | Replace with active test using current harness |
If the same ENGINE_GAP appears twice in one batch, stop and fix the shared support before continuing. Repeating an engine fix per card is a smell.
False-Green Signals
Treat as migration debt, not as healthy:
- Empty
.test.ts file.
- Test entirely commented
// LEGACY IMPLEMENTATION.
- Single keyword smoke check (Ward, Evasive, Bodyguard, Support).
- Client-side success without authoritative-state confirmation on hidden-zone, under-card, or chooser paths.
- Generated stub that compiles but still has
missingImplementation: true.
Decision tree (collapsed)
Card has text and no abilities? → Author (Step 4) after Probe
Card has abilities and failing test? → Triage: AUTHORING vs ENGINE_GAP
Card has empty/legacy test only? → TEST_SURFACE: write active test
Card claims missingImplementation: true? → Probe; if extension is bounded, do it
Same blocker hits 2+ cards in a batch? → Stop; fix shared engine support
Printed text is rules-ambiguous? → Ground via lorcana-rules first
Generation and Parser Tooling
The package contains parser and generator scripts. Inspect generated files when:
- A card looks machine-generated and structurally inconsistent with adjacent cards.
- Repeated malformed stubs suggest the generator, not the cards.
- Parser/text-shape normalization is the real blocker.
bun run --cwd packages/lorcana/lorcana-cards generate
bun run --cwd packages/lorcana/lorcana-cards generate-cards:all --skip-fetch
bun run --cwd packages/lorcana/lorcana-cards fetch
Do not paper over generator gaps with one-off card hacks. Either fix the generator or escalate.
Reusable Prompts
See PROMPTS.md.
Companion files
