PR-6 deck portability: export/import + standalone decrypt tool

[WilfordGrimley]

Summary

Implements docs/proposals/proposal-g-user-accounts-saved-decks.md's "PR-6, post-v1: deck portability" section (design-complete, owner-directed addendum) — the next un-built addendum in sequence after the 5 already-merged core saved-decks PRs. Does not implement PR-7 (art provenance), sequenced after this.

• Export: "Export my decks" on MyDecksPage downloads the account's complete encrypted bundle (every deck's ciphertext + wrapped DEK + nonces, both wrapped-master-key slots, salt/iterations, formatVersion: 1) as one JSON file. Works even while locked (no unlock required) — verified with a dedicated test, since it's the spec's own headline scenario ("a user who's forgotten their passphrase can still export").
• Import: ImportDeckModal decrypts a selected bundle using its own passphrase/recovery key (not necessarily the live session's — a bundle may come from a different account or instance), then re-encrypts every deck under the current session's master key and persists each as a brand-new row (key: null) — never overwrites by matching key or name, per spec.
• Revision tracking: DeckPayload v2 adds private revision/modifiedAt fields (never server-visible). Bumps the payload's own version 1→2 — read as the spec's "PR-6/PR-7 shared versioning rule" (see code comments/ledger note on that interpretation). A v1 legacy payload is upgraded forward on load, never rejected. buildDeckPayload's content shape deliberately excludes these bookkeeping fields so the existing dirty-check baseline isn't broken by an ever-changing modifiedAt.
• Standalone decrypt tool: decrypt-saved-deck-export/decrypt.mjs — zero npm dependencies (Node's own node:crypto WebCrypto only), cross-verified against a bundle actually produced by the browser's crypto code (not just its own mirrored logic). The trust anchor for "if this site vanishes tomorrow, your decks are still yours."
• No backend changes — export/import are fully served by the existing getSavedDecks/getCryptoProfile/saveDeck/saveCryptoProfile endpoints. makemigrations --check not applicable (no models touched).

Deviations / open items

1. MIT license on the standalone tool — this repo is GPL-3.0; decrypt-saved-deck-export/ is declared MIT, mirroring federation-hash-tool/'s existing precedent for a standalone tool meant to be freely reusable by forks. Flagged explicitly (code header, readme, docs/upstreaming/extractable-primitives.md note) rather than silently decided — owner call if this precedent shouldn't extend here.
2. version-as-shared-versioning-rule reading — the spec's prose on formatVersion vs. the private payload's version field is genuinely ambiguous in isolation; implemented as: EXPORT_FORMAT_VERSION (bundle's own public envelope version) vs. DeckPayloadV2.version (private per-deck version, doubling as the "PR-6/PR-7 shared counter"). Documented in code comments for the PR-7 implementer/owner to correct if this reading is wrong.
3. Import requires the current session already unlocked — importing into a brand-new account with no crypto profile yet isn't handled by ImportDeckModal; the entry point is simply disabled until unlocked. Not required by spec, a reasonable scope limit.
4. On-site Privacy Policy text (about.tsx) not touched — PR-6 doesn't collect/retain/share any new data (export only surfaces what the server already stores); judged not to need a policy-text change. Separately, the existing saved-decks feature (Discord OAuth + account data) isn't mentioned on that page at all — a pre-existing gap from the original 5-PR build, out of scope here, worth a follow-up.
5. Wiki: this changes what a signed-in user sees (new Export/Import buttons on My Decks) — Saved-Decks-adjacent wiki page needs a corresponding update at merge time (cloud-session checklist item per CLAUDE.md).

Test plan

• npx jest — 460 tests passing (includes new deckPayload.test.ts, deckExportImport.test.ts, ImportDeckModal.test.tsx, plus updated SaveDeckModal/LoadSafetyModal/MyDecksPage tests for revision-tracking and the new UI)
• npx tsc --noEmit — clean
• npx eslint — clean
• npx prettier@2.7.1 --check — clean (pre-commit hook also ran clean)
• npx playwright test tests/SavedDecks.spec.ts — 7/7 passing, including two new real-browser tests (an actual file download for Export, an actual <input type="file"> selection for Import) — screenshot of the shipped UI: see PR description image below
• node --test decrypt-saved-deck-export/tests/decrypt.test.mjs — 2/2 passing
• python3 .github/scripts/docs_lint.py — clean
• Manually cross-verified: a bundle built by the actual frontend crypto code (Jest, real WebCrypto) was decrypted successfully by the standalone Node CLI tool, confirming true cross-implementation compatibility, not just mirrored logic passing its own tests

Screenshot (My Decks page with Export/Import buttons, from the new Playwright test):

(screenshot captured locally at /tmp/pr6-my-decks-export-import.png during the Playwright run — not uploaded to this description; reviewer can reproduce via npx playwright test tests/SavedDecks.spec.ts -g "downloads a bundle".)

🤖 Generated with Claude Code

Merge-conflict resolution (master moved: #243 PR-5 share links, #244 proposal-h docs, #246 canary report)

• Merged origin/master into this branch (merge commit, no rebase) to resolve the 3-way textual conflict in docs/README.md, docs/features/saved-decks.md, and frontend/src/features/savedDecks/MyDecksPage.tsx — resolution kept both PR-5 (share links) and PR-6 (deck portability) content: docs/features/saved-decks.md now has both "Per-deck share links (PR-5)" and "Deck portability (PR-6)" sections, and MyDecksPage.tsx renders both the Import/Export controls and the per-deck Share modal.
• Semantic (non-textual) conflict found and fixed: frontend/src/features/savedDecks/deckShare.ts and SharedDeckViewer.tsx (PR-5, from master) typed a shared deck's decrypted payload as DeckPayloadV1. PR-6's deckPayload.ts changed parseDeckPayload to always upgrade to DeckPayloadV2 — these files didn't textually conflict during the merge but failed tsc --noEmit afterward. Fixed by retyping both to DeckPayloadV2 (what parseDeckPayload actually returns).
• Also fixed, in the same docs/README.md conflict resolution: an unrelated pre-existing regression where Add PR-5 per-deck share links (proposal-g §PR-5) #243's stale source branch had reverted the proposal-b-bleed-normalization.md status row from PARTIAL back to BUILDING (undoing docs: fix stale Proposal B status (PR-2/PR-3 shipped) #234's fix) — restored to PARTIAL since I was already touching that exact table row.

Re-verification after merge

• npx tsc --noEmit — clean (after the DeckPayloadV2 fix above)
• npx jest --silent — 55 suites / 470 tests passing (full suite, not just savedDecks)
• npx jest --silent src/features/savedDecks — 13 suites / 58 tests passing
• npx playwright test tests/SavedDecks.spec.ts --project=chromium — 7/7 passing
• npx prettier@2.7.1 --check on all files touched by the merge resolution — clean
• gh pr view 242 --json mergeable — now MERGEABLE (was CONFLICTING)

[WilfordGrimley]

Owner sign-off recorded: the standalone tool's MIT license (mirroring federation-hash-tool's precedent) is approved by the project owner, 2026-07-20, in the orchestration session directly — verbatim: "go for the license, mirror federation-hash-tool, if we end up making any changes to it we can gpl them." This resolves this PR's own "Deviations / open items" note #1 (MIT license — owner call). Recorded in the private orchestration ledger (proxyprints-orchestration/DECISIONS.md, 2026-07-20 entries) for cross-session verification. No longer an open item.