- Single-job workflow at .github/workflows/ci.yml (~49 lines including load-bearing comments)
- runs-on: ubuntu-latest, timeout-minutes: 10
- Uses actions/setup-node@v4 with cache: 'npm' (per RESEARCH CI Pitfall A — never cache node_modules/)
- Node 22 (per RESEARCH § Environment Availability)
- Triggers on push to main and pull_request to main
- Steps: checkout → setup-node → npm ci (lockfile-strict) → npm run ci (lint + test + validate-assets + build)
- Per CONTEXT user pushback: NO matrix, NO test reporters, NO Codecov, NO release automation
- Local npm run ci exits 0 (53 tests passing across 12 files); workflow will be green on push
- Structurally enforces every Phase 1 success criterion on every commit going forward
7 commits across 3 TDD tasks (RED + GREEN per task) + .gitkeep cleanup;
36 Vitest tests across 7 test files green; npm run build clean under
TypeScript strict; all 6 CORE requirements (CORE-04 through CORE-09)
covered by at least one assertion.
Key structural decision documented in SUMMARY: SaveDB is a single
common-contract interface, not a union of IDBPDatabase | LocalStorageDBAdapter.
The union shape failed TypeScript-strict at the build gate; the interface
refactor isolates the type-system cast to one location at openSaveDB().
src/save/ now contains 7 production files + 7 test files. The .gitkeep
firewall marker exists only to make empty directories trackable in git;
it can be retired once the directory has real content (per Plan 01-01
SUMMARY's pattern — 'firewall-as-directory pattern').
- codec.ts: exportToBase64 / importFromBase64 via lz-string with
MAX_IMPORT_BYTES=50MB DoS cap (T-01-02 in plan threat model); import
validates against SaveEnvelopeSchema before returning. lz-string sync
caveat documented per RESEARCH Pitfall 5 (Web Worker mitigation deferred
to Phase 8 per CONTEXT D-09)
- index.ts: 14 public re-exports — the only entry point Phase 2 should
import from. Includes the LocalStorageDBAdapter class so consumers can
type-check the fallback path explicitly if needed
[Rule 1 - Bug] Build was failing because the original SaveDB type was a
union (IDBPDatabase<SaveDBSchema> | LocalStorageDBAdapter) — TypeScript
cannot resolve method calls through a union when each branch has
differently-shaped overloads ('no compatible signature' on every db.put).
Fixed by:
- Defining SaveDB as a single common-contract interface that both
backends MUST satisfy (get/put/delete/getAll/transaction with
conditional-type RecordOf<S> return values)
- Hoisting the canonical SavedRecord/SnapshotRecord/StoreName types
into db-localstorage-adapter.ts (lower-level module) and re-exporting
them from db.ts to avoid a circular import
- Casting the idb-returned IDBPDatabase to SaveDB at the open-call
boundary (the casts are isolated to openSaveDB; Phase 2 only sees
the SaveDB interface)
- Promoting SnapshotEntry to a type-alias of SnapshotRecord so
snapshots.ts no longer redeclares the shape and can rely on
canonical types
Tests: 36/36 pass under 'npx vitest run src/save/' (full suite incl
sentinel: 37/37). 'npm run build' exits 0 under TypeScript strict.
'npm run lint' is not invoked here because Plan 02 (eslint-firewall) has
not landed yet — the lint script will fail until it does, by design per
the Plan 01-01 SUMMARY ('Plan 02 owns it').
SUMMARY documents the ESLint flat config, the 9 element types, the
single CORE-10 rule, the deliberate-violation fixture, the Vitest test
that runs ESLint programmatically against the violator, and the four
auto-fixed deviations (typescript-eslint parser-only integration,
real render target file for the violator import, eslint-import-resolver-
typescript wiring, tsconfig.app.json test-file exclusion).
Verifies: npm run lint -> 0 errors / 0 warnings; npm test -> 2/2 pass;
npm run build -> green; eslint --no-ignore on violator -> exits 1.
Self-Check: PASSED.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- round-trip.test.ts (3 tests): full pipeline EXPORT -> IMPORT -> MIGRATE
-> WRAP -> UNWRAP -> IDB PUT -> IDB GET exercising every save layer
file end-to-end (CORE-09 + CORE-04 + CORE-06 + CORE-07); plus DoS-cap
rejection at MAX_IMPORT_BYTES + 1; plus malformed-Base64 rejection
RED phase per TDD plan-level gate. Tests fail because codec.ts does not
exist yet.
- db.ts: openSaveDB() opens IndexedDB ('tlg-save', v1) with two object
stores (saves singleton + save_snapshots keyed); on openDB rejection
(private mode, blocked, quota exceeded) falls back to LocalStorageDBAdapter
per CORE-04 contract
- db-localstorage-adapter.ts: ~110-LoC adapter exposing the same minimal
get/put/delete/getAll/transaction surface as idb's IDBPDatabase, namespaced
under tlg.saves.<id> and tlg.save_snapshots.<id>; transaction() shim
proxies straight through (localStorage has no real transactions)
- snapshots.ts: snapshot(envelope) writes to save_snapshots and prunes to
RETAIN=3 newest by savedAt descending (CORE-08); listSnapshots() returns
newest-first; entropy suffix on snapshot IDs avoids same-ms collisions
- persist.ts: requestPersistence() returns {granted, apiAvailable} for all
4 navigator.storage scenarios per CORE-05 + RESEARCH Pitfall 2
Test infra fixes: snapshots.test.ts and db.test.ts cannot deleteDatabase
between tests because openSaveDB leaves an open connection that idb caches
(deleteDatabase blocks indefinitely). beforeEach instead clears store
contents directly. The fallback test calls vi.resetModules() BEFORE
vi.doMock('idb') so the freshly-imported db.ts picks up the rejecting
openDB stub, and re-imports LocalStorageDBAdapter from the same module
graph so instanceof checks against the same class identity.
Tests: 12/12 pass (npx vitest run src/save/db.test.ts
src/save/snapshots.test.ts src/save/persist.test.ts).
Full save suite: 33/33 pass (Task 1 + Task 2 combined).
TypeScript-strict; no 'any' in production code (CLAUDE.md).
- src/sim/__test_violation__/violator.ts deliberately imports from
src/render/__firewall_target__.ts to trigger the firewall rule.
- src/sim/__test_violation__/lint-firewall.test.ts runs ESLint
programmatically (with ignore: false) against the violator and
asserts boundaries/element-types fires with severity=error and the
message mentions both 'sim' and 'render'.
- src/render/__firewall_target__.ts is a minimal export so the
boundaries plugin can resolve the import to a real path on disk.
Without a real target, the plugin marks the import as isUnknown
and silently skips the rule (verified empirically; see SUMMARY).
- eslint.config.js gains an import/resolver: typescript block so the
TS-aware resolver follows extension-less imports
('../../render/foo' -> src/render/foo.ts). Required by the
boundaries plugin's element classification of import targets.
- tsconfig.app.json excludes *.test.ts and src/sim/__test_violation__/
so 'tsc -b' does not try to typecheck Node-API-using test code with
the DOM-only project's lib settings; vitest still discovers them
via its own include glob.
- Added eslint-import-resolver-typescript as devDep.
Verifies green:
npm run lint -> 0 errors, 0 warnings (violator excluded)
npm test -> 2/2 pass (sentinel + firewall)
npm run build -> tsc -b clean, vite build clean
npx eslint --no-ignore src/sim/__test_violation__/violator.ts
-> exits 1 with the expected
boundaries/element-types error
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Documents the asset-provenance gate that landed (validator script, Zod sidecar
schema with the 6 CLAUDE.md fields + optional provenance_schema_version, refused-
sample PNG, tmpdir-isolated Vitest enforcement test) and the resume protocol for
Task 2 (Path A AI-generate / Path B hand-painted-or-licensed-photograph fallback /
Path C defer with explicit IOU). Per plan's autonomous: false flag and orchestrator
spawn instructions, the human curates the 10–20 north-star reference images.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- .planning/season-7-end-state.md answers principle-level the three questions
per CONTEXT D-08: (a) what rest state means, (b) what the finite Roothold
ceiling is tied to (count of authored fragments + Seasons), (c) the coda's
tonal register (warm/quiet/specific/final). Cites SEAS-04, SEAS-09, SEAS-10,
STRY-08; ROADMAP Phase 7; PITFALLS #1.
- Includes the explicit 'What this document is NOT' boundary section so
treatment-level scope creep is structurally rejected (binary-choice scene
text, ending paragraph text, Lura's final line, credits screen — all
authored Phase 7, not Phase 1).
- scripts/doctrine.test.ts is the only automated enforcement of both Phase-1
doctrine docs (per CONTEXT D-07: no UX-string lint rule). Asserts file
existence + required H2 sections + required source citations + boundary
disclaimer. 8 assertions / 2 doc files.
- vitest.config.ts include glob extended to scripts/**/*.test.ts so
doctrine.test.ts is discovered by 'npm test'.
- tsconfig.node.json include extended to scripts/**/*.ts so the strict-TS
gate covers the new doc-lint test alongside the existing build configs.
- 'npm test' green: 2 test files, 9 tests passing (sentinel + doctrine).
- Per CONTEXT D-09: lives in .planning/, not docs/.
Rule 3 [Blocking]: vitest.config.ts and tsconfig.node.json include globs
extended to discover the new TypeScript test file (existing globs only
covered .mjs scripts and src/ tests).
- db.test.ts (4 tests): IDB-primary path opens both stores + round-trips
saves and save_snapshots; localStorage-fallback path via vi.doMock('idb')
asserts LocalStorageDBAdapter is returned and tlg.saves.main is written
- snapshots.test.ts (4 tests): basic put + listSnapshots, empty store
returns [], CORE-08 5-then-3 retention with newest-first ordering, and
pruned entries are oldest by savedAt
- persist.test.ts (4 tests): all 4 navigator.storage scenarios per
CORE-05 + RESEARCH Pitfall 2 (granted true / false / throws / missing)
RED phase per TDD plan-level gate. Tests fail because db.ts / snapshots.ts /
persist.ts / db-localstorage-adapter.ts do not exist yet.
- 2 happy-path tests: empty globs, valid YAML round-trip
- 3 throw assertions covering the schema-violation matrix:
* numeric id (violates stable-string-ID rule)
* season out of [0,7] range
* Markdown frontmatter missing required id
- All 5 tests pass; full Phase-1 suite remains green
- Proves the throws that fail npm run build at module-eval time
- scripts/validate-assets.mjs: walks ASSETS_DIR (default 'assets'), requires every
non-sidecar non-.gitkeep non-README file to carry a sibling <name>.provenance.json
validating against Zod ProvenanceSchema (6 required fields per CLAUDE.md / AEST-08
+ optional provenance_schema_version per RESEARCH Open Question #2). Excludes
assets/__samples__/refused/ so the proof-of-gate fixture passes the gate.
- assets/__samples__/refused/no-provenance.png: 1x1 transparent PNG with no sidecar;
the gate-proof artifact per CONTEXT D-03.
- scripts/validate-assets.test.ts: Vitest integration test covering both cases.
Positive: real /assets/ tree must exit 0. Negative: per-test-run mkdtemp under
os.tmpdir() with one orphan PNG; runs validator with ASSETS_DIR pointing at the
tmpdir; asserts exit 1 + clear error message + cleanup in afterAll. No risk of
polluting the real /assets/ tree (BLOCKER 2 fix).
- vitest.config.ts: extend include glob to also pick up scripts/**/*.test.ts (Rule 3
blocking fix — without this the new test file is invisible to vitest).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- New eslint.config.js (flat, ESLint 9) declaring 9 element types:
the seven Phase-1 firewall directories (sim, render, ui, save,
content, audio, store) plus the template's app + game.
- One rule, severity error: sim cannot import from render or ui (CORE-10).
- Default posture allow — Phase 1 enforces ONE rule, not closed-by-default.
- src/sim/__test_violation__/ excluded from default lint glob; the rule's
end-to-end correctness is proven by Task 2's Vitest test, not by 'lint
exits 0 on clean code'.
- Added typescript-eslint as devDep (parser only — no rule sets) so
ESLint can parse .ts/.tsx (Espree default cannot). Documented as a
Plan 02 deviation in 01-02-SUMMARY.md (Rule 3 — Blocking).
Verifies green on the clean codebase: 0 errors, 0 warnings via
'npm run lint'. Stderr notices from boundaries plugin about deprecated
rule name (element-types vs dependencies in v6) and legacy selector
syntax are informational only — they don't count as ESLint warnings.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Banned Mechanics table enumerates 17 banned patterns (gacha, lootboxes,
narrative gating, daily login, streaks, limited-time, energy/stamina,
rewarded ads, push notifications, loss-aversion copy, countdown timers,
Season skipping, time-skip purchases, hint systems, mobile nag UX, etc.)
- Allowed Engagement section names the 4 affordances that respect presence
rather than demand it (Memory Storm opt-in, while-you-were-away letter,
tab-title bloom indicator, Season-transition save-export reminder)
- Review Checklist provides 3 questions for every UX/copy/monetization change
- Source Documents section cites PROJECT.md, REQUIREMENTS.md, CLAUDE.md,
.planning/research/PITFALLS.md
- Per CONTEXT D-07: doctrine is enforced by review, NOT by lint rule on UX
strings (the doc explicitly notes this and proposes no lint rule)
- Per CONTEXT D-09: lives in .planning/ alongside other internal design docs
Wave 1 of Phase 1 complete. Phaser 4 + React 19 + Vite 8 + TypeScript 6
scaffold builds (npm run build green); 15 Phase-1 deps installed at
locked versions; 7 architectural-firewall directories ready under src/;
repo-root /content/ + /assets/ trees ready; Vitest (happy-dom) +
Playwright wired with passing sentinel; package.json scripts
pre-declared for the entire Phase 1 plan-set so Wave 2 can run in
parallel without colliding on package.json.
3 deviations auto-fixed (1 blocking, 2 missing-critical):
1. Built scaffold by hand because @phaserjs/create-game v1.3.2 is
interactive-only — plan's documented fallback path was used.
2. build script wraps tsc -b before vite build so strict-TS gates
every build (CLAUDE.md Code Style invariant).
3. Added *.tsbuildinfo to .gitignore (TS 6 incremental cache files).
Wave 2 readiness: Plan 02 must use ESLint 9 flat-config format
(eslint.config.js); legacy .eslintrc.* not supported. fake-indexeddb
pre-installed for Plan 03 IDB tests. inkjs + inklecate installed but
no .ink files compiled (Phase 2 PIPE-02 owns that).
- vitest.config.ts: happy-dom environment (so Plan 03's IndexedDB tests
can layer fake-indexeddb on top of happy-dom's window per RESEARCH);
passWithNoTests:false enforces RESEARCH CI Pitfall B (a green CI run
must mean tests *ran*, not 'no tests existed'); include glob covers
src/**/*.test.ts(x) and scripts/**/*.test.mjs.
- playwright.config.ts: testDir 'tests/e2e' (not yet created — first spec
lands in Phase 2 PIPE-07); webServer config wires npm run dev so smoke
tests can self-start the dev server; baseURL pinned to vite default.
- src/__sentinel__.test.ts: a single test asserting 1+1===2 AND that
globalThis.window exists, proving the runner is wired and happy-dom is
active. To be deleted once real tests exist (Plan 03 onward).
- npm test → 1 file, 1 test passed in 593ms.
- npx playwright --version → Version 1.59.1 (matches RESEARCH lock).
- Built equivalent React + Vite + TypeScript scaffold by hand because the official
npm create @phaserjs/game@latest scaffolder is interactive-only and the documented
--template/--yes flags are ignored (verified 2026-05-08 with create-game v1.3.2).
Plan Step 1 explicitly authorizes this fallback. Resulting tree mirrors the
official template shape: index.html, src/main.tsx, src/App.tsx, src/PhaserGame.tsx,
src/game/main.ts, src/game/scenes/Boot.ts.
- Installed Phase-1 production deps at versions verified in RESEARCH.md:
phaser@4.1.0, react@19.2.6, react-dom@19.2.6, idb@8.0.3, lz-string@1.5.0,
zod@4.4.3, crc-32@1.2.2, gray-matter@4.0.3, yaml@2.8.4, inkjs@2.4.0.
- Installed Phase-1 dev deps: vite@8.0.11, @vitejs/plugin-react@6.0.1,
typescript@6.0.3, @types/react@19, @types/react-dom@19, @types/node@22,
vitest@4.1.5, @vitest/ui, happy-dom, fake-indexeddb@6 (for Plan 03 IDB tests),
@playwright/test@1.59.1, eslint@9, eslint-plugin-boundaries@6.0.2, inklecate@1.8.1.
- Created the seven architectural-firewall directories under src/ with .gitkeep
markers (sim, render, ui, save, content, audio, store) — siblings to the
template-provided src/game/ — so Plan 02's ESLint boundaries rule has clean
targets per CLAUDE.md 'Architectural Firewall'.
- Created repo-root /content/ (with /dialogue/ and /seasons/ subdirs) and /assets/
trees per CONTEXT D-11, D-12.
- Pre-declared all downstream-required scripts in package.json so Plans 02–06 only
edit code, not script keys: dev, build, preview, lint (--max-warnings 0 per
RESEARCH CI Pitfall C), test (--passWithNoTests=false per CI Pitfall B),
test:watch, validate:assets, compile:ink (no-op stub for Phase 1; Phase 2
replaces with real inklecate invocation), ci.
- TypeScript strict mode enforced via tsconfig.json + tsconfig.app.json + tsconfig.node.json.
- npm run build succeeds (tsc -b && vite build) producing dist/index.html and
dist/assets/index-*.js (~1.5MB Phaser bundle; code-splitting deferred to Phase 2+
when actual scenes exist).
User locked four implementation decisions:
- AI asset pipeline: minimum-viable schema + sidecar provenance + CI gate;
vendor/model deferred to Phase 5; 10–20 hand-curated AI generations as
Phase 1 north-stars
- Save v1: minimal payload (Phase 2 fields only); synthetic v0→v1 migration
proves the chain works; first real migration ships in Phase 4
- Doctrine docs: anti-FOMO consolidation + Season 7 principle-level rest-state
contract; both in .planning/; no CI/lint enforcement
- Phase 1 scaffold caps at the 5 success criteria — BigQty, Zustand store, and
tick scheduler defer to Phase 2
Pushback recorded: user prefers minimum-viable infrastructure for support
systems; no ceremonial workflows.
josh