Refresh checkpoint: profile metadata publishing and pictures

CHECKPOINT-encryption.md

@@ -1,4 +1,4 @@
-# Checkpoint — All audit items closed (2026-08-21)
+# Checkpoint — Profile metadata publishing + profile pictures (2026-08-22)
 
 A stopping point you can return to if this session is closed. Everything below was
 verified green at the moment this file was written.
@@ -6,177 +6,97 @@ verified green at the moment this file was written.
 ## Where things are
 
 - Project: `/home/avi/Projects/0_Nostr`
-- Git repo: `master` @ `f7db29e` ("Add SSRF guard, signer queue cap, secret echo, request
+- Git repo: `master` @ `a6329d5` ("Publish profile metadata (name + picture) so external
-  timeout"). Before it: `130d7e2` (zeroization), `4461307` (owner-only writes), `4bd7660`
+  clients show it"). Before it: `db81f8d` (profile deletion with undo), `9a8f334`
-  (legacy vault perms), `4d4dfde` (CSP + navigation guards), `6e627a3` (upload tokens),
+  (audit checkpoint refresh), and the 2026-08-21 audit-fix commits (`f7db29e`, `d90b6e5`,
-  `4bde395` (IPC allowlist).
+  `130d7e2`).
 - Working tree is **clean** apart from this checkpoint update, which is committed right after.
 
-## What was completed: the full 2026-08-21 security audit remediation
+## What was completed
 
-All ten findings from the 2026-08-21 security audit are fixed.
+**Problem:** profiles created in the app never published a Nostr kind 0 metadata event,
+so other clients showed generated petnames ("evil iguana", "homeless leech") or a
+truncated npub instead of the user's chosen name.
 
-**#1 IPC method allowlist (`4bde395`, `main.ts` only):**
+1. **Automatic metadata on creation** — `create_profile` now publishes a kind 0 event
-- `RENDERER_METHODS` set in the Electron main process: the three native methods
+   with the label as `name`/`display_name` to all enabled relays (best-effort; relay
-  (`pick_image`, `link_preview`, `upload_image`) plus every Rust-backend method used by
+   failures never block creation).
-  `frontend/src/lib/api.ts`.
+2. **"Publish name" action for existing profiles** — new button on every Profiles-screen
-- `isAllowedMethod()` gate at the top of the `backend:request` handler; unknown methods get
+   card plus CLI `publish-name <npub>`. Returns a per-relay report shown in the UI.
-  `{status:'error', code:'unknown_method'}`, are logged, and never reach the backend.
+3. **Profile pictures end-to-end** —
-- No behaviour change for the real UI (only `api.ts` calls `window.backend.request`).
+   - Vault: optional `picture: Option<String>` per profile (serde default → old vaults
+     load unchanged).
+   - Backend: `set_profile_picture` validates http(s) URLs only, stores the URL, and
+     publishes kind 0 including `picture`; clearing supported (`None`).
+   - GUI: "Picture" button opens a modal — paste a URL, upload a file via the existing
+     nostr.build pipeline, or remove; avatar shows the picture everywhere in-app.
+   - CLI: `set-picture <npub> <url>`.
+4. **Nested-runtime safety** — metadata publishing runs on a dedicated OS thread with its
+   own tokio runtime, so both sync (CLI) and async (IPC server) callers are safe.
+5. **Finished prior session's delete/undo work** — exposed `undo_history` in
+   `AppStateView`, fixed invalid Button variants / missing icon / null-safety errors so
+   the frontend typechecks again.
+6. **Test hygiene fix** — `Settings::default()` points at real relays and tests were
+   silently publishing events to them (one got rate-limited by damus.io). All test suites
+   now use offline settings; test time dropped from ~126 s to ~3 s.
 
-**#2 Upload pick tokens (`6e627a3`):**
+## Commits added in this session
-The renderer used to send raw filesystem paths back to main for upload, so a compromised
-page could read+publish arbitrary local files to nostr.build. Now:
-- `pickImage()` (`main.ts`) mints a random 32-hex-char token per picked file via
-  `crypto.randomBytes`; tokens map to the file record in `pickedTokens` (main-process memory
-  only). The renderer receives `{token, name, mime}` — it never sees any path.
-- `upload_image` accepts only a valid `token`; the token is consumed (deleted) before the
-  upload starts, so each pick authorises exactly one upload. Unknown/used/expired tokens
-  return `{status:'error', code:'unknown_token', message:'That image selection has expired.
-  Please attach it again.'}` — the user just re-picks the image.
-- Renderer updates: `PickedImage.path` → `PickedImage.token` (`types.ts`), `api.uploadImage(token)`
-  (`api.ts`), context type + callback (`AppProvider.tsx`), `ComposeScreen.onAttach` uses
-  `image.token`.
-- Fake backend mirrors the contract: `upload_image` without a non-empty `token` throws an
-  `unknown_token` error, so tests exercise the same protocol rule.
 
-**#3 Header-based CSP + navigation/window guards (`4d4dfde`):**
+- `a6329d5` Publish profile metadata (name + picture) so external clients show it
-- The static CSP meta tag was **removed** from `frontend/index.html` and replaced with
-  response headers stamped by `main.ts` (`onHeadersReceived`, mainFrame only):
-  - `CSP_PROD` for `app://` pages: `script-src 'self'` — **no `'unsafe-inline'`**, so an
-    injected `<script>` cannot execute. `connect-src 'self'` (all real network goes through
-    the backend/main process); img-src keeps `https:` for remote feed/preview images.
-  - `CSP_DEV` when `NOSTR_GUI_DEV_URL` matches: keeps `'unsafe-inline'` (Vite's React-refresh
-    preamble is an inline script) but pins `connect-src` to localhost instead of the old
-    wide-open `ws:` — HMR still works.
-  - The meta had to go because it would also have blocked the dev preamble; headers let us
-    ship strict-prod / workable-dev from one HTML file.
-- `will-navigate`: any navigation away from app:// or the dev origin is blocked; http(s)
-  targets open in the system browser via `shell.openExternal`.
-- `setWindowOpenHandler`: all `target="_blank"` popups are denied; external links go to the
-  system browser. Unknown schemes are denied without opening anything.
 
-**#4 Legacy vault permissions (`4bd7660`):**
+## Verification commands run (all green)
-- The original CLI left `profiles_vault.json` files with default (often group-readable)
-  permissions containing plaintext keys. Now:
-  - On **every** startup, `load_vault()` calls the new `harden_stray_legacy_vaults()`
-    (`src/vault.rs`): each known legacy location is checked, and any file that parses as a
-    populated vault is chmod'd to 0600 — whether or not migration ever runs.
-  - During migration itself, the legacy file is tightened before the backup copy is made.
-  - Hardening is best-effort (`let _ =`) so odd filesystems can never break vault loading;
-    unparsable and keyless files are deliberately not touched.
-- One-time local cleanup done by hand in this session: repo-root `profiles_vault.json`
-  was chmod'd from 0664 → 0600.
-- New tests: recognised legacy vault tightened to 0600; unparsable file ignored; empty /
-  encrypted-but-populated vaults classified correctly.
 
-**#5 Owner-only-from-first-byte writes (`4461307`):**
+Rust (repo root):
-- `write_restricted` (`src/vault.rs`): the tmp file is now created with `.mode(0o600)`
-  (OpenOptionsExt), so it never exists under default umask permissions. The explicit
-  `set_permissions` stays, covering a tmp file left over by a crashed earlier run (where
-  `mode` would not apply to an existing file).
-- `backup_file`: replaced `fs::copy` — which gives the new file the *source's* permission
-  bits, so backing up a group-readable legacy vault briefly produced a fully readable copy —
-  with a manual open-at-0600 + `io::copy`, making backups owner-only from their first byte.
-- New test: backup of a 0664 source comes out 0600 with identical content.
-
-**#6 Key material zeroization (`130d7e2`, adds `zeroize = "1"`):**
-Decrypted secrets and derived keys no longer linger in unscrubbed heap memory:
-- `crypto.rs`: `decrypt_secret` returns `Zeroizing<String>` (self-shredding on drop); the
-  UTF-8 error path wipes the raw bytes too; a failed Argon2 derivation wipes its key buffer.
-- `profiles.rs`: `resolve_secret_key` / `resolve_active_secret_key` now return
-  `Zeroizing<String>`, so every transient plaintext key flowing to publish/upload/signer is
-  wiped when its scope ends. Newly generated keys in `create_profile` are wrapped the same
-  way. `reveal_secret_key` still hands a display copy to the UI by design.
-- `app.rs`: `lock()` and a new `Drop for App` zeroize the session vault key; wrong-password
-  derivations wipe their throwaway keys; password change wipes the previous vault key and
-  the old unlock key; `remove_password` moves decrypted values into storage via
-  `std::mem::take` without extra copies (plaintext-at-rest is that feature's purpose).
-- Callers in `publish.rs`, `uploads.rs`, `signer.rs` needed no changes (deref coercion).
-
-**Smaller items (`f7db29e`, main.ts + signer.rs):**
-- **SSRF guard**: `fetchLinkPreview` resolves the URL's host and refuses loopback/private/
-  link-local targets — literal IPs, DNS answers, and localhost/.local names all checked
-  (`ipv4IsPrivate`/`ipv6IsPrivate`/`resolvesToPrivateAddress`). Crafted note links can no
-  longer make the app probe localhost or the LAN.
-- **Signer queue cap** (`MAX_PENDING_APPROVALS = 20`): a relay flooding `sign_event`
-  requests cannot grow the approval queue unboundedly or bury a genuine prompt; overflow
-  requests get the standard "no decision" error.
-- **NIP-46 secret echo**: when the nostrconnect:// link carried a secret, the client's
-  `connect` request must echo it back or it is refused ("did not include the expected
-  secret"), proving the link arrived unmodified.
-- **Backend request timeout**: `backendRequest` reaps any round-trip after 120s (generous
-  enough for multi-relay publishes), deleting its pending entry so a hung backend cannot
-  leak promises.
-
-## Commits
-
-- `f7db29e` "Add SSRF guard, signer queue cap, secret echo, request timeout" — main.ts +
-  signer.rs (+204/−3).
-- `130d7e2` "Zeroize transient secret key material in memory" — Cargo.toml, crypto.rs,
-  profiles.rs, app.rs (+77/−23).
-- `4461307` "Create vault files owner-only from the first byte" — src/vault.rs (+48/−2).
-- `4bd7660` "Tighten permissions on leftover legacy vault files" — src/vault.rs (+79/−3).
-- `4d4dfde` "Enforce header-based CSP and block window open/navigation" — main.ts + index.html.
-- `6e627a3` "Replace upload file paths with single-use pick tokens" — main.ts, api.ts,
-  types.ts, AppProvider.tsx, ComposeScreen.tsx, fakeBackend.ts (+62/−27).
-- `4bde395` "Restrict renderer IPC to an explicit method allowlist" — main.ts (+49 lines).
-- `dafed33` "Refresh checkpoint with IPC allowlist hardening".
-
-## How it was verified (all green)
 
 ```
-cargo test            # 84 passed
+cargo test            # 96 passed; 0 failed
-cargo clippy --all-targets   # clean
+cargo clippy --all-targets   # 0 errors/warnings from session code
 cargo fmt --check     # clean
-cargo build --release # ok
+cargo build --release # success
-npm run typecheck     # clean (frontend/)
-npm run lint          # clean (pre-existing module-type warning only)
-npm run format:check  # clean
-npm test              # 78 passed (14 files)
-npm run electron:build # compiles the Electron main process
-npm run build         # rebuilds the React bundle (dist/)
 ```
 
-Manual protocol checks worth doing once:
+Frontend (`frontend/`):
-- DevTools console: `window.backend.request('not_a_method')` → `unknown_method` envelope (#1).
-- DevTools console: `window.backend.request('upload_image', {path:'/etc/passwd'})` →
-  `unknown_token` error; no file is read (#2). Real attachment flow works unchanged.
-- DevTools console: `document.cookie` / injected `<script>` does not run; Application tab
-  shows the CSP header on the app:// document (#3). Link previews open in the system browser.
 
-## How to resume
+```
+npm test              # 14 files, 78 tests passed
+npm run typecheck     # clean
+npm run lint          # 0 errors
+npm run format:check  # clean
+npm run build         # vite build success
+npm run electron:build  # tsc electron main success
+```
 
-1. Open the repo: `cd /home/avi/Projects/0_Nostr`
+## How to use / reproduce
-2. State is committed: `git status` should be clean; `git log --oneline -4` shows `6e627a3`
-   on top.
-3. Launch as usual: `cd frontend && npm start` (backend already built in `target/release/`).
-4. Re-run verification with the commands above.
 
-## Outstanding / next steps (if you continue)
+GUI:
 
-Nothing outstanding from the audit — all ten findings are closed:
+```bash
+cd ~/Projects/0_Nostr/frontend && npm start
+```
 
-| # | Finding | Fix commit |
+- New profiles publish their name automatically on creation.
-|---|---------|-----------|
+- Existing profiles: **Profiles → Publish name** button, or **Picture** button to set a
-| 1 | Unrestricted renderer→backend IPC | `4bde395` |
+  photo (URL paste or file upload) which publishes immediately.
-| 2 | Arbitrary file upload paths | `6e627a3` |
-| 3 | CSP `unsafe-inline`, no nav guards | `4d4dfde` |
-| 4 | Legacy vault world-readable | `4bd7660` |
-| 5 | Permission race windows on write | `4461307` |
-| 6 | Key material not zeroized | `130d7e2` |
-| 7 | Link-preview SSRF | `f7db29e` |
-| 8 | Signer queue flooding | `f7db29e` |
-| 9 | NIP-46 secret not verified | `f7db29e` |
-| 10 | No backend request timeout | `f7db29e` |
 
-**Profile deletion with undo functionality** (2026-08-22):
+CLI:
-- Users can delete a profile via `nostr-manager-backend delete-profile <npub>`, which moves the profile to an in-memory undo stack rather than permanently removing it
-- Profiles can be restored with `nostr-manager-backend undo-delete`, which pops the last deleted profile from the undo stack and re-adds it to the vault (becoming active if no other profile exists)
-- The undo stack is in-memory only (lost on process exit); a persistent implementation would require vault metadata changes
-- Vault must be unlocked or unencrypted for deletion to be permitted
-- Existing profile data and vault integrity are preserved
-- CLI commands: `delete-profile <npub>` and `undo-delete`
-- Rust changes verified: `cargo test` (91 tests), `cargo clippy --all-targets` clean, `cargo fmt --check` clean, `cargo build --release` successful
 
-Possible future work (not audit items): DNS-rebinding TOCTOU in the SSRF guard, OS keyring integration for the vault password, an automated dependency-audit CI job (`cargo audit`, `npm audit`), GUI integration for profile deletion/undo in the ProfilesScreen.
+```bash
+B=~/Projects/0_Nostr/target/release/nostr-manager-backend
+$B list                                   # show profiles
+$B relays enable wss://nos.lol            # enable at least one relay first
+$B publish-name <npub>                    # republish stored name
+$B set-picture <npub> https://…/img.png   # set + publish picture
+```
+
+Verified live during the session: kind 0 events confirmed present on nos.lol,
+relay.primal.net, relay.damus.io and the user's own wss://nostr.l484.com via direct
+websocket queries; Iris shows the stored names after publish.
+
+## Notes & next steps
+
+- Nostr has no relay-to-relay sync: names/pictures are only visible on relays they were
+  published to. Keep major relays enabled so clients that don't read your own relay can
+  see the profile.
+- Clients cache profiles; hard-refresh (Ctrl+Shift+R) after republishing.
+- Possible follow-ups: rename profiles (edit label + republish), NIP-05 identifiers,
+  banner/about fields in the edit UI.