phase-rs · matthewevans · Aug 19, 2026 · Aug 19, 2026
@@ -20,7 +20,7 @@ If you find yourself about to type `pub enum Foo { ... NewVariant ... }` in `cra
 
 ### Stage 1: Existence verification (5-grep protocol)
 
-The variant might already exist under a different engine-native name. The mtgish AST and the engine vocabulary are not 1:1; the engine often has the concept under a different name.
+The variant might already exist under a different engine-native name. External data formats and the engine vocabulary are not 1:1; the engine often has the concept under a different name.
 
 > The canonical engine surface is `data/engine-inventory.json` (gitignored). Run `cargo engine-inventory` to (re)generate it locally before grepping for existing variants.
 
@@ -46,7 +46,7 @@ rg -n "<concept_keyword>" crates/engine/src/database/synthesis.rs
 **Stage 1 verdicts:**
 
 - **EXISTS_SAME_NAME**: variant exists. Stop. Wire to it.
-- **EXISTS_DIFFERENT_NAME**: concept exists under engine-native name (e.g., mtgish `CreateTriggerUntil` ↔ engine `Effect::CreateDelayedTrigger`). Stop. Map mtgish AST to the existing slot.
+- **EXISTS_DIFFERENT_NAME**: concept exists under an engine-native name (e.g., an external `CreateTriggerUntil` maps to engine `Effect::CreateDelayedTrigger`). Stop. Map the source data to the existing slot.
 - **EXISTS_AS_PARAMETER**: concept exists as a parameter value of a more general variant (e.g., "untapped" exists as `Tap { negated: true }`). Stop. Use the parameter form.
 - **DOES_NOT_EXIST**: proceed to Stage 2.
 
@@ -131,7 +131,7 @@ If you've made it through all three stages with EXTEND_OK / WITHIN_SECTION verdi
 ## Anti-patterns this skill prevents
 
 - "Audit said add variant X" → adding sibling without verification (the audit is wrong about a third of the time per session metrics)
-- "Engine doesn't have it" → without grepping for engine-native names (CreateTriggerUntil mtgish ↔ CreateDelayedTrigger engine)
+- "Engine doesn't have it" → without grepping for engine-native names (external CreateTriggerUntil ↔ CreateDelayedTrigger engine)
 - "Just one more sibling" → ignoring the sibling-cluster smell and compounding parameterization debt
 - "Unify under one type" → crossing CR rule sections and conflating runtime resolvers
 - "Type-only stub returning a placeholder true/false" → masks runtime correctness; pair with real evaluation logic before shipping. The historical example: `ReplacementCondition::CastViaKicker` initially shipped with a `=> true` stub that silently over-applied to non-kicked spells. Resolved by adding `Option<KickerVariant>` to the variant and tracking `SpellContext.kickers_paid` at cast resolution — the runtime now actually evaluates the gate. Always verify the resolver does what the variant doc-comment claims.
@@ -150,7 +150,6 @@ Refusing is the correct outcome when any stage fails. Coverage waits, architectu
 ## Related artifacts
 
 - Workspace `CLAUDE.md` — "Parameterize, don't proliferate" principle (the policy this skill enforces)
-- Crate `mtgish-import/CLAUDE.md` Rule §8 — audit-verdict filter discipline
 - Memory note `feedback_parameterize_dont_proliferate.md` — user directive recording this as a hard rule
 
 ## Inputs / outputs

@@ -448,7 +448,7 @@ Do NOT start implementation — do not run `$engine-implementer` — until a ful
 
 Adding a new enum variant or a new field to a struct-variant has compile-time blast radius across the WHOLE workspace. This was the single most common gap class across multi-round plan reviews — make it an explicit plan section and a mandatory reviewer check.
 
-- **New field on a struct-variant** (e.g. `split` on `Effect::SearchLibrary`): `#[serde(default)]` does NOT make it optional for Rust *literals*. Every `Effect::Variant { … }` construction across ALL crates (engine, phase-ai, mtgish-import, tests) must set it; every non-`..` destructure must bind it or add `..`. Enumerate with `grep -rn "Effect::Variant {" crates/` — treat the grep, not a hand-list, as authoritative.
+- **New field on a struct-variant** (e.g. `split` on `Effect::SearchLibrary`): `#[serde(default)]` does NOT make it optional for Rust *literals*. Every `Effect::Variant { … }` construction across ALL crates and tests must set it; every non-`..` destructure must bind it or add `..`. Enumerate with `grep -rn "Effect::Variant {" crates/` — treat the grep, not a hand-list, as authoritative.
 - **New `WaitingFor` / enum variant**: breaks every EXHAUSTIVE `match` (no wildcard). Enumerate ALL sites by grepping a recently-added sibling variant (`rg "WaitingFor::SomeRecentVariant"`), then classify each match as exhaustive (needs an arm) vs `_`/`..`/`if let`/`matches!` (doesn't). Span engine + phase-ai + server-core. One `WaitingFor` variant typically needs ~5 arms (e.g. `acting_player`, `ai_support/candidates`, `phase-ai/decision_kind::classify`, `phase-ai/search::fallback_action`, `game/scenario::waiting_for_kind`).
 - **THE SILENT KILLER — `matches!` dispatch gates**: a handler reachable only if its variant is listed in a `matches!` guard (e.g. `engine_resolution_choices::handles`) is NOT compile-checked. Omit it and the handler is silent dead code — the action falls through to InvalidAction; clippy and the type system say nothing. Grep for `matches!`/membership predicates that gate dispatch and add the variant; add a test asserting the action is *handled* (not InvalidAction).
 

@@ -7,8 +7,6 @@ description: "End-to-end phase.rs implementation pipeline: plan, review-plan, im
 
 This is the orchestrator for the phase.rs implementation pipeline. It runs as a **skill in the main thread** so it can spawn agents for every step that benefits from fresh context (plan review, surgical implementation, implementation review). Do not turn this into an agent — agents cannot spawn sub-agents, which is what made earlier versions silently degrade.
 
-> **⚠️ `mtgish` is dormant — DO NOT route implementation work through it.** `mtgish/`, `crates/mtgish-import/`, and `data/mtgish-*` are NOT live consumers of the engine, parser, or card data. Reject any plan section, executor edit, or review fix that touches mtgish files; surface it to the user instead of silently shipping it. PRs that only modify mtgish are rejected on sight.
-
 ## Roles
 
 | Step | Where it runs | Why |

@@ -9,8 +9,6 @@ Produce an implementation plan for the phase.rs engine. Design for the class, no
 
 This skill produces the plan only. The plan-review loop belongs to the caller — when invoked from `/engine-implementer`, the orchestrator owns the loop. When invoked standalone, run `/review-engine-plan` against the plan yourself and iterate until clean.
 
-> **⚠️ `mtgish` is dormant — out of scope for ALL plans.** Never plan changes to `mtgish/`, `crates/mtgish-import/`, or `data/mtgish-*`. The import pipeline is not a live consumer of the engine or parser; new variants, parser patterns, and effects do NOT need to be mirrored there. If a task description references mtgish, surface the contradiction and stop — do not silently include mtgish in the plan.
-
 ## Input
 
 A task description: parser enhancement/fix, or engine mechanic enhancement/fix. May reference cards, Oracle text patterns, CR rules, or coverage gaps.

@@ -1,9 +1,9 @@
 # MSH Wave 3 — Replacement Effects (Cluster B): Implementation Plan
 
 Generated by Wave3Planner (via /engine-planner). Decision LOCKED: **Approach (A)** —
-no new `ReplacementDefinition` fields (mtgish is out of bounds). Reuse the existing
-`execute` field + a new `TokenCoreTypeMatches` condition variant + a new
-`Effect::RemoveAllDamage` primitive. Zero mtgish edits.
+no new `ReplacementDefinition` fields. Reuse the existing `execute` field +
+a new `TokenCoreTypeMatches` condition variant + a new
+`Effect::RemoveAllDamage` primitive.
 
 > REVISION (round 2): All sections passed review EXCEPT the Wolverine combat-batch
 > simultaneity case (previously deferred as a follow-up — that was rules-incorrect).
@@ -80,7 +80,7 @@ parser limits in `parse_damage_modification_static` (oracle_replacement.rs:4014;
   value(AttachedTo, tag("enchanted creature"))))`, consulted by `parse_damage_source_subject_filter` (:4401)
   before `parse_type_phrase`.
 Result: `ReplacementDefinition::new(DamageDone).damage_modification(Double).damage_source_filter(AttachedTo)`.
-NO type-surface changes, NO mtgish impact. CR: 614.1a, 301.5/702.6 (Equipment/equip → AttachedTo host link).
+NO type-surface changes. CR: 614.1a, 301.5/702.6 (Equipment/equip → AttachedTo host link).
 
 **⚠ CLASSIFICATION GUARD (reviewer nit #4 — BLOCKING for the FIRST ordering).** The "parser-only / lowest-risk /
 implement FIRST" classification rests entirely on `damage_source_filter = AttachedTo` resolving as a *damage-source*
@@ -114,8 +114,8 @@ Approach (A) — no new ReplacementDefinition field:
   (fn head replacement.rs:3283; match :3291; add the new arm in the condition-match body — the brief's earlier
   ":3624" pointer is one arm region inside that same match, both are the same match block) against
   `spec.characteristics.core_types`. Parser sets `condition(TokenCoreTypeMatches{core_types:[Creature]})`.
-  Run `/add-engine-variant` gate. mtgish-SAFE: mtgish-import only CONSTRUCTS ReplacementCondition variants,
-  never matches them exhaustively; engine's `evaluate_replacement_condition` is the only exhaustive match (add one arm).
+  Run `/add-engine-variant` gate. Engine's `evaluate_replacement_condition`
+  is the only exhaustive match (add one arm).
   AUDIT all other `ReplacementCondition::` match sites for exhaustiveness before finalizing.
 - Parameterize check: only 2 token-characteristic gates (subtype, core-type) → below 3-variant threshold, distinct
   fields → NOT a sibling-cluster smell.
@@ -157,19 +157,11 @@ Approach (A) implementation:
   the Phase-B-before-Phase-C invariant (combat_damage.rs:869-944 / deal_damage.rs:446) that makes the heal
   combat-batch-correct. Standard `/add-engine-effect` (resolver in effects/mod.rs, EffectKind, targeting, AI/FE
   serialization, coverage).
-- **mtgish-safety (CITATION FIX #1, CONFIRMED):** adding `Effect::RemoveAllDamage` cannot break mtgish. mtgish-import
-  only ever CONSTRUCTS engine `Effect` / `ReplacementCondition` values — it never exhaustively matches the engine
-  `Effect` enum. Where it does match over `Effect`, every site has a catch-all: `rewrite_bound_x_in_effect`
-  (convert/action.rs:451), `rewrite_any_target_filter_in_effect` (:597), `apply_player_target` (:5458, a named
-  `other =>` strict-fail arm), and `apply_token_flags` (:6459) — all fall through on unrecognized variants. (The
-  previously-cited `convert/token.rs:607` was WRONG: that line is a `match effect { TokenCopyEffect::AddAbility… }`
-  over `TokenCopyEffect`, an mtgish AST type, NOT engine `Effect` — removed.) Conclusion unchanged: mtgish-safe.
-  Re-confirm these catch-alls compile during impl (engine-only build), but no mtgish edits are needed or permitted.
 CR: 614.1a, 510.2 (combat damage simultaneous — same-batch instances must not heal each other),
 120.6 (marked damage remains until cleanup — heal removes early), 120.1/120.3, 614.5.
 
 ## Sequential implementation order (all share oracle_replacement.rs → strictly sequential)
-1. Mjölnir — parser-only, no type surface, no mtgish, lowest risk. FIRST **after the classification-guard
+1. Mjölnir — parser-only, no type surface, lowest risk. FIRST **after the classification-guard
    confirmation test passes**; if AttachedTo-as-source-filter needs runtime wiring, reclassify and reorder
    (do not block 2/3).
 2. Divine Visitation — `TokenCoreTypeMatches` condition + token-substitution shape/applier (execute-reuse). MEDIUM.

@@ -2,7 +2,6 @@
 
 **Scope:** 3 cards on the `main` checkout, one coherent wave.
 **Cards:** Dragon Man, Reformed Robot (Cluster C, CDA) · Wolverine, Claws Out (Cluster C, combat-damage assignment) · Ms. Marvel, Elastic Ally (Cluster H, new FilterProp).
-**mtgish:** OUT OF BOUNDS. No step in this plan touches `mtgish/`, `crates/mtgish-import/`, or `data/mtgish-*`. No mtgish mirroring is required for any of these changes.
 **Skills applied:** `/add-static-ability`, `/add-trigger`, `/add-engine-variant`, `/oracle-parser`.
 
 ---
@@ -260,4 +259,4 @@ Only `oracle_static/` is touched by two cards, in **different files/functions**
 3. **Dedicated vs parameterized self-comparison (Ms. Marvel).** Plan recommends the dedicated `PowerExceedsBase` (2 siblings < threshold; mirrors `ToughnessGTPower`; avoids a multi-site refactor under concurrent edits). Reviewer may mandate `PtSelfComparison` instead — explicit decision point; categorical boundary (CR 208) holds either way.
 4. **`Aggregate` doc-comment** says "battlefield objects" but the resolver is zone-general — correct it as part of Dragon Man (prevents the next agent repeating the recon error).
 5. **Inert-until-regen.** All three are parser changes; `from_oracle_text` unit tests pass even with stale deployed data. Tests MUST use `add_real_card` + rehydrate to exercise the real card-data path; regenerate card-data before claiming "supported".
-6. **No stop-and-return blockers.** No mtgish dependency, no missing infrastructure — all three implementable on `main` now.
+6. **No stop-and-return blockers.** No missing infrastructure — all three are implementable on `main` now.
@@ -223,7 +223,6 @@ Runtime tests use `add_real_card` + rehydrate and drive the REAL declare-attacke
 
 ## Constraints honored
 - Do NOT touch `StaticMode` (#4010/#3958) — scope rides `attack_defended`; ✓.
-- mtgish dormant — not referenced; ✓.
 - Surgical/additive; no whole-file rewrites; commit by pathspec; Tilt-first (no raw cargo); ✓.
 - All parser dispatch via nom combinators; ✓ (Nom Compliance section).
 - No bool flags — `AttackTargetFilter` + `RestrictionPlayerScope` + `RestrictionExpiry` typed enums; ✓.

@@ -183,7 +183,7 @@ The **parser is the detector** throughout.
 
 ## 3. Step-by-Step Implementation Plan
 
-> Verification cadence per CLAUDE.md: `cargo fmt --all` directly; everything else via Tilt (`tilt logs clippy/test-engine`, `./scripts/tilt-wait.sh`). Do NOT run `cargo build/clippy/test` directly. mtgish is out of scope — touch nothing under `mtgish/`, `crates/mtgish-import/`, `data/mtgish-*`.
+> Verification cadence per CLAUDE.md: `cargo fmt --all` directly; everything else via Tilt (`tilt logs clippy/test-engine`, `./scripts/tilt-wait.sh`). Do NOT run `cargo build/clippy/test` directly.
 
 ### Step 1 — Add `AbilityTag::PowerUp` + `keyword_str()` (gate (a))
 - **`types/ability.rs:11645`** — add to `AbilityTag` (after `Backup`): `/// CR 602.5b + CR 602.1: This ability originated from a Power-up keyword definition.\nPowerUp,`. (Re-grep `602.5b` and `602.1` before writing — verified `MagicCompRules.txt:2541`, `:2510`.)

@@ -267,9 +267,8 @@ predicate (the existing honest-refusal path, oracle_casting.rs:244-255); the Swa
 `Condition_If` detector continues to flag it. We add a guard test asserting **no** unconditional
 flash grant leaks. We do **not** ship a green-but-wrong flash grant.
 
-**mtgish-safety statement.** Zero edits to `mtgish/`, `crates/mtgish-import/`, or `data/mtgish-*`.
 No new `Effect`, `AbilityCondition`, `AdditionalCostOrigin`, or `SpecialClause` variant (all
-already exist), so no mtgish mirroring is implicated. Changes are confined to parser files
+already exist). Changes are confined to parser files
 (`conditions.rs`, `lower.rs`, `oracle_modal.rs`, `oracle_nom/condition.rs`) plus tests.
 
 ### Blast-radius bounding (mandatory — EMH + We Say Thee Nay! fixes)

@@ -309,7 +309,7 @@ One coherent change set; nothing half-wired:
 13. (If Quantum Reduction flash shipped) `types/ability.rs` `ParsedCondition::AdditionalCostPaid` + its flash-timing evaluation.
 14. Integration tests per §11; snapshot updates if any card's parsed abilities change (`cargo coverage` to confirm the 17 cards drop their unsupported markers).
 
-**Verification cadence (Tilt-first):** `cargo fmt --all` (direct, always) → if Tilt up, `./scripts/tilt-wait.sh --timeout 240 clippy test-engine card-data`; else direct `cargo clippy --all-targets -- -D warnings` + `cargo test -p engine`. `cargo coverage` (direct one-shot) to confirm the 17 cards. `pnpm type-check`/`pnpm lint` for the FE diff. Do NOT touch `mtgish/`, `crates/mtgish-import/`, or `data/mtgish-*` (dormant — out of scope).
+**Verification cadence (Tilt-first):** `cargo fmt --all` (direct, always) → if Tilt up, `./scripts/tilt-wait.sh --timeout 240 clippy test-engine card-data`; else direct `cargo clippy --all-targets -- -D warnings` + `cargo test -p engine`. `cargo coverage` (direct one-shot) to confirm the 17 cards. `pnpm type-check`/`pnpm lint` for the FE diff.
 
 ---
 

@@ -30,7 +30,7 @@ Add a permanent CI regression gate proving the `StaticModePresence` O(1) index k
 
 ## Applicable skills
 
-No new engine surface is introduced (no new effect, keyword, trigger, static, variant, or parser pattern). This is a **test-only** change composing existing pub APIs, so `/add-engine-effect`, `/add-engine-variant`, and the parser skills do not apply. The governing guidance is `/card-test`'s foot-gun list (adapted: these are enumeration/perf tests, not cast-pipeline tests) and the CLAUDE.md verification-matrix contract. No `mtgish` involvement.
+No new engine surface is introduced (no new effect, keyword, trigger, static, variant, or parser pattern). This is a **test-only** change composing existing pub APIs, so `/add-engine-effect`, `/add-engine-variant`, and the parser skills do not apply. The governing guidance is `/card-test`'s foot-gun list (adapted: these are enumeration/perf tests, not cast-pipeline tests) and the CLAUDE.md verification-matrix contract.
 
 ## Analogous trace (hard gate)
 

@@ -512,4 +512,4 @@ compile error, so there is no silent gap).
 
 No engine files (`casting_costs.rs` untouched); no `can_afford_x_at_least_one` primitive (deferred,
 justified); no changes to `x_reference.rs` / `x_value.rs` semantics; no display-derive revival in sim
-mode; no other policy's behavior; `mtgish` untouched.
+mode; no other policy's behavior.
@@ -31,7 +31,6 @@ reviews:
     - "!client/public/**.jpg"             # binary assets
     - "!client/src/wasm/**"               # generated by ./scripts/build-wasm.sh
     - "!**/*.d.ts"                         # committed wasm type artifact
-    - "!data/mtgish-cards.json"           # generated import data
     - "!data/semantic-audit.json"         # generated by `cargo semantic-audit`
     - "!data/parser-gaps.json"            # generated by `cargo parser-gaps`
     - "!data/engine-inventory.json"       # generated by `cargo engine-inventory`

@@ -174,7 +174,7 @@ jobs:
         # caught before it lands on the branch that feeds staging/preview.
         env:
           PROPTEST_CASES: ${{ github.event_name == 'pull_request' && '32' || '256' }}
-        run: cargo nextest run --profile ci --partition count:${{ matrix.shard }}/4 --workspace --exclude phase-tauri --exclude mtgish-import --features engine/proptest --status-level fail --final-status-level fail
+        run: cargo nextest run --profile ci --partition count:${{ matrix.shard }}/4 --workspace --exclude phase-tauri --features engine/proptest --status-level fail --final-status-level fail
 
   card-data-gate:
     name: Card data (generate, validate, coverage)

@@ -71,7 +71,6 @@ client/coverage/
 data/*
 !data/standard-cards.txt
 !data/mtgjson/
-!data/mtgish-cards.json
 data/mtgjson/AtomicCards.json
 data/mtgjson/CardTypes.json
 data/mtgjson/Meta.json
@@ -119,10 +118,6 @@ parser-batch-plan.md
 parser-fallback-plan.md
 tmp/
 
-# External reference / scratch
-mtgish/
-mtgish*.tgz
-
 # Python bytecode cache
 __pycache__/
 *.pyc

@@ -52,7 +52,6 @@ tilt up             # continuous build/test — leave running
 ## Pull requests
 
 - Target `origin/main` (`phase-rs/phase`).
-- Don't modify `mtgish/`, `crates/mtgish-import/`, or `data/mtgish-*` (dormant);
   PRs that only touch them are rejected.
 - If you used an LLM, use `.github/PULL_REQUEST_TEMPLATE.md` for the PR body,
   fill every section, and report the model on its canonical `Model:` line; see