Files
vardkey/ai-note.md
T

39 KiB
Raw Blame History

AI Agent Note — VardKey

Short working handover for the next model. Keep it short.

  • TODO.md (Norwegian) is the requirement specification: §1–§18 are binding and describe the app as it is now, §19–§32 are history as one line each, and §32 holds only what is left. The full reasoning is in the git history.
  • The binding rules live in TODO.md and README.md. This note does not repeat them — it carries the handover, the rules that are easy to get wrong, and the traps that already cost time.
  • HANDOVER.md and ai-thought.md were deleted as outdated. Do not recreate either.

Status

  • Done and verified: everything through §32.7, plus §32.8 stages A, B and C. Schema version 13. 532 green tests. The StartOS package is built, sideloaded and running on a StartOS 0.4.0 box; the package version is still 0.1.0:0.
  • What is left, in this order: §32.8 point 7 (the user's manual end-to-end pass — see below), then §32.6 (the documentation drift), then §32.5 (the raise to 1.0.0:0 with §18 step 12). 32.6 has to come after the rounds it documents, and 32.5 last.
  • A round stops when it is done. Do not start the next item unless the user asks.
  • Verified how: the automated suite, the user's manual browser testing, and the install and test round on a StartOS 0.4.0 box. The user runs the suite; the agent has no shell.
  • Tests: python3 -m unittest discover -s tests. They run on a Windows copy at C:\Users\<brukarnamn>\Downloads\helsesjekk, so they must pass there as well as on Linux.

§32.8 point 7 — the one open item

The user's manual end-to-end pass (BR-003). It covers: first setup and the mandatory admin step; login and logout; an expired or invalid session; a direct API call without a session; one full check and the CSV export; another threshold and key count; deleting all data with the confirmation; restart, backup and restore; PWA offline and online, also after logout; plus the six finished points (future date rejected, sticky top bar, the login lockout, the admin task, a rotation that does not change older history, and software closures in the history and the CSV).

Run so far, on the box: everything above is OK except three things.

  • The required task does not hold the service back. Set Admin Password shows on Tasks as recommended, so the user can start the app without an admin. That is a deviation from §32.8 point 1, which asks for a task the user cannot skip. The decision is open: try a stricter severity in startos/init/requireAdmin.ts and re-test on the box, or accept the deviation and document what actually happens in §32.6. SDK 3.0.4 has no TaskCondition for "the admin password is unset".
  • The CSRF test is not run yet: a POST to /api/settings/wallet-software with the session cookie but without the X-CSRF-Token header must answer 403, and must not store anything.
  • PWA offline and online after logout is not confirmed yet.

One expectation in the run was wrong, and the app is right: a key with a software-sourced software_incompatible is not blocked in the key select, because that alert carries no key_id. Do not "fix" that.

The wallet-secret guard (round 32-B)

secret_guard.py is the one place that decides whether a piece of text looks like a wallet secret. It is local (no network call), silent (the text never reaches a log, a message or an exception) and narrow (it matches only the shapes that are unmistakable, so an ordinary note is not refused).

  • What it recognises: a PEM private-key block; an extended private key with any SLIP-132 prefix (xprv, yprv, zprv, tprv, uprv, vprv) at 100 base58 characters or more; a WIF key (51 or 52 base58 characters starting 5, 9, K, L or c); a BIP38 key (6P, 58 characters); and a run of 12 or more consecutive BIP39 words, also in the middle of a sentence. It returns a short English label such as "a seed phrase", never the value, and rejection_message(field, label) never repeats it either.
  • The wordlist is embedded, 2048 words, from bitcoin/bips bip-0039/english.txt. tests/test_secret_guard.py checks it: 2048 words, alphabetical, 3–8 lower-case letters, and no two words sharing their first four letters. That last check is the strong one. Do not edit the list without that test.
  • Where it is wired in: validation.py calls _reject_secret() from the four text helpers _text, _key_name, _model and _required_path, so every free-text field is covered by construction. auth.py checks the admin username the same way; the password is deliberately not checked, because a long passphrase is a legitimate password.
  • The refusal is a ValidationError (SecretDetected), so the existing handlers answer 400. app.py needed no change at all — do not add a handler for this. A plain ValueError would have been caught by STORAGE_ERRORS and answered 500.
  • The guard runs before the field's length limit. A seed phrase pasted into a 60-character label field is answered with "looks like a seed phrase", not with "too long". This order is deliberate and pinned by tests; do not swap it back.
  • The reminder in the UI is one line, in the key setup — a single p.secret-warning above the whole key list in static/setup.html, and nowhere else. The page notice is a different thing and stays on health-check.html, setup.html and settings.html. The server side is unchanged.
  • The limitations are documented in the module docstring and are not to be played down: a bare 64-character hex string is not refused (it is indistinguishable from a transaction id); only the English wordlist is used; a secret split across several fields, or in a shape outside the patterns, is not caught; and a false positive is possible for a note that happens to be twelve or more BIP39 words in a row.

Alerts, closures and key identity — the parts that are easy to get wrong

The rules are in TODO.md §7; these are the traps that have already cost time.

  • The acknowledgement stores the key. alert_acknowledgements.key_id is nullable and holds the key the alert belonged to, or NULL for a software alert raised by the settings list. An alert with a key is acknowledged and hidden by (code, key_id); an alert without one stays on (code, source). This closed a real bug: a rename — or a rotate with keep_name=false — used to reopen a closed alert, because source carries the name the key has now.
  • A key-sourced security alert carries Key N: <name>, and the row decides. build_alerts() reads the key checks into a lookup by key id, and _security_alerts() looks the row's own key_id up in it. When the row carries the checked key, the key is the source and the alert belongs to it, so it is closed by a rotation; otherwise the source is the plain software name and it is closed by removing that software from the settings. The provider name is never compared with the names that are configured now, so a rename or a rotation cannot turn a key alert into a software one. rotation_alert_targets(), software_alert_targets() and newest_alert_check_id() all read the same build_alerts(), so they see the same source, which is what keeps an alert closable.
  • The contact alert belongs to the checked key. It gets key_id and source = Key N: <name>, and its message says "this key". The severity stays info and the action stays contact_updated. It blocks its key like every other key alert, because blocking_alert() and the blockingAlerts filter in the form only look at key_id.
  • software_incompatible has two forms. A software from the settings list keeps source = the software name, key_id = None, actions [remove_software], and carries software. A wallet software that is the name of a configured cloud key can never be removed from the settings, so that form belongs to the checked key instead: key_id = the checked key, source = Key N: <name>, actions [replace_hardware], and no software value, so a settings removal can never close it. Only that form names the software in its message, because its source is the key.
  • alerts.py still needs the cloud key names, but only for software_incompatible. build_alerts() and collect_alerts() take cloud_names=(), threaded through rotation_alert_targets(), software_alert_targets(), and the four call sites in app.py (_send_alerts, _blocking_alert, _rotate_key, _close_removed_software), plus newest_alert_check_id(). app.cloud_key_names(keys) builds the tuple. If newest_alert_check_id() does not get exactly the same names as collect_alerts(), the alert's source differs, the acknowledgement stores check_id = None, _is_acknowledged() refuses to hide an alert without a check id, and the alert can never be closed. Keep the default () on build_alerts(). _send_health_check_list() passes the same tuple for its needs_follow_up field.
  • Every closure of a check is shown, key-sourced or not. app.closed_alerts() and database._grouped_closures() list every acknowledgement that carries a check_id, so a rotation, a replaced hardware, an updated contact, an ignored key alert, and a software alert closed by removing it are all shown. GET /api/health-checks/<id> returns them as closed_alerts, and the CSV has one last column, Follow-up closed, with code: action (YYYY-MM-DD), hanging on the check it closed through check_id. The history table shows Alert, Action and Closed only — the source is not a column.
  • The key card shows the next due date with the alert note. createKeyCard() in static/app.js always renders Next check due and adds the alert note on top. The note comes in addition to the date, never instead of it.
  • The type is part of a key's identity. The setup locks it, the settings never change it, and only a rotation may pick the other one: POST /api/keys/<id>/rotate takes an optional type, and leaving it out keeps the current type. Since schema 13 a rotation does not write the key row: it inserts a new row and retires the old one.
  • A key that becomes a cloud key must not take a taken name. A cloud key offers its own name as a wallet software choice, so a name the list already holds is refused with 409 through WalletSoftwareNameInUse — a ValidationError subclass, so it has to be caught before the general ValidationError in _rotate_key, or the 409 turns into a 400. validate_rotation() therefore takes the wallet software list as a fourth argument.
  • A rename needs no rotation and no confirmation. In the rotation section the name can be changed on its own through POST /api/settings/keys, for any key, and the address path of a key with no saved check can be changed there the same way. The single button is Save keys.

Deliberately left out

remove_software is still a dead end in two places: a software-sourced security_vulnerable offers it next to ignore (so the alert stays closable, but one button leads nowhere), and a software-sourced security_compromised offers it as its only action, so an alert whose "software" is a cloud key name can never be closed. Neither is fixed.

Rules that are easy to get wrong — do not re-open

  • One submission records one key. There is no "round" grouping, and each key's next check is its own date plus one calendar year, computed on the server and not editable in the form.
  • A key that could not be checked records nothing else. That covers available = 0 and a recorded address path that does not match. wallet_software and ai_security_checks must be empty; a non-empty value is a 400, not something dropped silently. Nothing is stored for them, and software_compatibility_status becomes NOT_ASSESSED.
  • Availability is required and offers only Yes/No. A cloud key collects only Availability; its compatibility is derived from it, and its own name is the wallet software.
  • A hardware key needs an explicit compatibility status, and Hardware works needs an explicit Yes or No. A missing value, "" and NOT_ASSESSED are all a 400.
  • The contact confirmation is required for every submission, including one for a key that could not be checked, and it has no "not assessed" value.
  • NOT_ASSESSED is not offered or accepted anywhere else.
  • Address path verified is never a form field again. The server derives it from the comparison.
  • The address path is locked after the first saved check, and no two keys may share one — in the setup, in the settings or in a rotation. The comparison is canonical: m/48h/0h/0h/2h and m/48'/0'/0'/2' are the same path, and the stored spelling is never rewritten.
  • The type is locked by the setup, and only a rotation may change it. A rename alone is not a rotation.
  • An alert is closed by the user, not by a newer check. A newer check that raises nothing does not clear an older alert, and a key with an open alert cannot get a new check at all.
  • An acknowledgement stores the action name, not a free reason. rotate has its own endpoint, because it requires a new address path.
  • Every alert message says what is wrong, which key or software, and what closes it, and it does not repeat the name in source — except the cloud-key form of software_incompatible. See §7.
  • The export is plain CSV only. XLSX is not built and the app does not promise Excel compatibility. The column names and the column order are an agreement; do not change them. A text cell that a spreadsheet would read as a formula is neutralized with a leading apostrophe, and that is the only change a cell may receive.
  • The PWA cache stores only a valid shell answer. isCacheable() in static/sw.js requires response.ok, response.type === "basic" and !response.redirected. The allow-list is the other half: the worker never names /api/*, /health, /login or /robots.txt.
  • No page is meant to be crawled. All six pages carry <meta name="robots" content="noindex, nofollow"> in <head>, _send_security_headers() in app.py adds X-Robots-Tag: noindex, nofollow, and static/robots.txt answers Disallow: /. static/robots.txt must not be added to the service worker's shell list.
  • The navigation lives in the top bar, not on the page. static/session.js builds the menu, the panel and the .topbar-actions wrapper on the five application pages; /login has none. The dashboard markup carries no href for /health-check, /settings, /history or /api/export, and the export link and its note are on the history page. The page the user is already on is a <span> with aria-current="page", not a link.
  • The Attention card is the only summary card. Its detail line carries the overdue list when something is overdue and the earliest due date otherwise; .overview-grid is one column at every width.
  • An empty alert list renders nothing. The count beside the heading says No alerts.
  • The two summary fields are additive and mean what they say. key_names follows key_numbers and carries the name the check was recorded with, read from the snapshot on the check; needs_follow_up is a property of the saved check, so it does not change when an alert is closed or a key is renamed. Do not "fix" it to follow collect_alerts(), which hides acknowledged alerts and would rewrite history.
  • The prohibition lists in sections 1 and 13 of TODO.md stay unticked. They are standing prohibitions, to be re-confirmed before release — with the one amendment for the installable PWA.
  • Whether a date is overdue is decided in the browser, so /api/key-status stays independent of the current date.
  • /api/previous-values still returns a security map that nothing consumes any more, and its address_path per key is no longer used by the form, which pre-fills the path from keys.address_path. Both are deliberate.
  • delete_all_data() deletes the checks before the keys. key_checks.key_id is ON DELETE RESTRICT, so the other order fails. The acknowledgements are deleted explicitly, because they have no foreign key at all. The delete has to clear every row of keys, active and retired alike.

Data model (schema version 13)

  • keys: key_number (1–multisig_key_count), name, type (hardware/cloud), model, address_path, active, rotated_at, created_at. There is no device column. The name is editable in the settings; the type and the address path are the identity of the key, the path must be unique across the active rows, and the type only changes through a rotation. model is optional documentation and is not locked.
  • health_checks: checked_at, next_check_due, wallet_software, software_version, software_compatibility_status, software_comments, contact_info_verified, comments, plus the legacy sparrow_version / nunchuk_version.
  • key_checks: one row per submission — key_id, key_name, key_type, key_model, available, hardware_works, firmware_version, seed_backup_verified, address_path_verified, address_path, comments. The three snapshot columns are a frozen copy of the key the check was recorded for.
  • ai_security_checks: provider (free text), key_id (nullable), status, comments, with UNIQUE (health_check_id, provider).
  • wallet_software: name, created_at. Seeded with DEFAULT_WALLET_SOFTWARE (Sparrow Wallet) when the list is empty, never emptied through the API.
  • alert_acknowledgements: code, source, key_id, check_id, reason, acknowledged_at.
  • app_settings: name (PRIMARY KEY), value. Holds multisig_threshold, multisig_key_count, admin_username, admin_password_hash, session_secret, failed_login_count and failed_login_until.
  • tool_checks no longer exists.
  • Startup migration rebuilds keys when it still has device, the old key_number CHECK, or no active flag; key_checks when it uses the old layout or still holds the two cloud columns; and health_checks when its CHECK constraints still allow the old values. It then drops tool_checks if it is there, adds the newer columns through OPTIONAL_COLUMNS, backfills the snapshot on every check from the key it belongs to, and creates the partial unique index over the active row. The UNIQUE on keys.key_number is now a partial unique index over the active row (idx_keys_active_key_number … WHERE active = 1).
  • The table definitions are {table} templates, so a rebuild can create the same table under a temporary name. health_checks is a parent table, so the whole migration runs with PRAGMA foreign_keys = OFF, checks PRAGMA foreign_key_check before committing, and rolls back on any failure.
  • tests/test_database.py carries LEGACY_SCHEMA, SCHEMA_VERSION_5, SCHEMA_VERSION_8, SCHEMA_VERSION_9, SCHEMA_VERSION_10, SCHEMA_VERSION_11 and SCHEMA_VERSION_12 as literal older schemas. Copy that pattern for a new migration test: write the old schema, run initialize_database(), then assert on the migrated rows.
  • There is no data migration for an older test database: the user deletes the data folder and configures the setup again. An acknowledgement written before version 11 has no key, and the older contact alert used the source Contact information, which no longer matches, so the alert it closed is raised again.

API

Pages: /, /setup, /settings, /health-check, /history, /login. Endpoints: GET /health, GET /api/session, POST /api/admin, POST /api/login, POST /api/logout, GET /api/setup, POST /api/setup, GET /api/keys, GET /api/key-status, GET /api/previous-values, GET /api/settings, POST /api/settings/keys, POST /api/settings/wallet-software, POST /api/keys/<id>/rotate, GET /api/health-checks, GET /api/health-checks/<id>, GET /api/alerts, POST /api/alerts/acknowledge, GET /api/export, POST /api/health-checks. Field details are in docs/application.md — which is behind the code, see §32.6.

Maintenance commands, run inside the container by the StartOS actions and usable by hand: maintenance.py delete-all-data and maintenance.py set-admin-password. Neither is an HTTP endpoint.

Five things that are easy to break:

  • New static files must be added to STATIC_FILES in app.py, or they return 404. That includes /settings, /settings.js, /login, /login.js, /session.js, /sw.js, /manifest.webmanifest, /icon.svg, the two raster icons /icon-192.png and /icon-512.png, and /robots.txt, which is text/plain; charset=utf-8 and is deliberately absent from the service worker's shell list.
  • /api/setup, /api/settings, /api/alerts, /api/key-status, /api/previous-values and /api/session are extra endpoints that are not in the minimum list in section 11.
  • STORAGE_ERRORS in app.py includes RuntimeError, because list_keys() raises it for a partial setup. Keep it there, or a broken database answers with a traceback instead of a 500. It also includes ValueError, so validate_admin_credentials() has to be called in its own try and mapped to a 400 before the storage handler — and it is why a refused secret has to be a ValidationError rather than a bare ValueError.
  • The settings, rotation and acknowledgement endpoints keep the same error handling as the existing ones: a bad payload is a 400, an unknown key or alert is a 404, a locked or duplicate address path is a 409, and a broken database is a 500. DuplicateAddressPath and WalletSoftwareNameInUse must be caught before the general ValidationError — in _save_key_settings and _rotate_key respectively — or the 409 turns into a 400. A missing or invalid session is a 401, a missing CSRF token is a 403, and a locked login is a 429.
  • newest_alert_check_id() in app.py derives the check id for the acknowledgement, the rotation and the software removal, so the same comparison rule is used in all three places. blocking_alert() is the module-level helper the 409 is built on, and _blocking_alert() reads the key id straight from the payload, so the rule applies before validation.

The rotate payload is replaced + keep_name + name + type + address_path, and type is optional. A rotation inserts a new row and retires the old one, and the acknowledgements are written against the old key id. GET /api/health-checks/<id> also returns closed_alerts, and each of its key_checks rows carries model as well. The #history-detail anchor and the ?id= query on /history are client-side only: they never reach the server, so no route changed for them.

The StartOS package — the parts that bite

The package is one service: one daemon, one container, one SQLite database on the persistent main volume, mounted at /data. No Postgres, no Redis, no external backend, no extra container.

  • The application and the package share one repository, at the root. A deliberate deviation from the guide's "a package repo is not a fork of the application": the rule assumes a published upstream image or a git submodule, and VardKey has neither. AGENTS.md states the deviation so nobody adds an upstream-project/ submodule to "fix" it.
  • Anything the Dockerfile does not COPY is not in the image. The modules go in with COPY *.py ./ and the assets with COPY static ./static; .dockerignore keeps tests, data, node_modules, javascript and the build output out of the context. static/ must be copied: app.py answers /health from code, but every page is read from ROOT/static, so an image without it reports a healthy service whose whole UI answers 404.
  • Three things must agree. startos/utils.ts holds uiPort (8080); startos/main.ts passes it as PORT and uses it in the readiness URL; startos/interfaces.ts binds it. The readiness check is checkWebUrl against http://127.0.0.1:8080/health, and /health must stay open. A red ECONNREFUSED 127.0.0.1:8080 blip before VardKey listening on port 8080 is just the first poll landing before Python has bound the socket; nothing to fix.
  • Two actions, registered through Actions.of() in startos/actions/index.ts, with their strings in startos/i18n/dictionaries/: deleteAllData.ts (delete-all-data) and setAdminPassword.ts (set-admin-password). Both run maintenance.py in a withTemp subcontainer, behind a confirmation the handler enforces. The generated password never travels into the container: maintenance.py generates it there and prints it.
  • The required task from §32.8 point 1 lives in startos/init/requireAdmin.ts and keeps the same property: no password travels from the host into the container. It raises and clears the task itself, because SDK 3.0.4 has no condition for "the admin password is unset". The probe fails open. Its behaviour on the box is still open — see §32.8 point 7 above.
  • Backups. startos/backups.ts is sdk.Backups.ofVolumes('main'). StartOS stops the service for the duration of a backup, so the database file is quiescent while it is copied — no snapshot hook is needed.
  • No CI. The guide assumes GitHub Actions; this repository is on Gitea, so there are no .github/workflows/ and there must not be any. The .s9pk is built locally with make.
  • The skeleton was written to match the Start9 template rather than generated by s9pk init-workspace, because that command was not available where it was written. The successful build and install confirm they are equivalent; @start9labs/start-sdk is pinned to 3.0.4. node_modules/ is not in the repository.
  • The root icon.svg is the package artwork, read from the repository root by the build. It is the same retro drawing as static/icon.svg, and it must stay under 40 KiB.
  • Documentation layout. README.md is the StartOS package README, with the guide's fixed heading set. The application's own reference is docs/application.md, which instructions.md links to. AGENTS.md is the repo-specific agent context, and CLAUDE.md is the one-line @AGENTS.md import. Both README.md and docs/application.md are behind the code — that is §32.6.

Traps that already cost time — do not repeat

  • with sqlite3.connect(...) does not close the connection. It only commits or rolls back the transaction. On Windows the database file stays locked and TemporaryDirectory.cleanup() fails with WinError 32. Always close explicitly with try/finally; see write_schema() in tests/test_database.py.
  • A module the image is missing is a runtime error, not a build error. auth.py was once left out of the Dockerfile, and both app.py and maintenance.py then died with ModuleNotFoundError inside the container while every test stayed green. The COPY *.py ./ line is what stops that recurring. secret_guard.py is covered by the same line.
  • Every server test module needs a session, and a redirect makes a missing one look like a pass. A page fetched without a session answers 302 → /login, and urllib follows it, so an assertion on page markup reads the login page's HTML instead of failing with 401. Use tests/server_session.py's SessionMixin. The manifest, the icons and /robots.txt are open while the pages are not.
  • urlopen raises HTTPError on a 4xx. A test that expects a 400 has to catch it and read the body from the exception; tests/test_secret_guard.py's post_status() helper is the pattern. A 429 is read the same way.
  • A field has a length limit, and the guard runs before it. Use a WIF key (52 characters) or a BIP38 key (58) when a test needs a secret that fits a short field, and the seed phrase where the field has room.
  • A required control inside a display:none container still blocks submission. Hiding a section is not enough: the required flags must be cleared. That is why renderSecurityChecks() shows an empty note instead of cards when a key could not be checked, why syncUncheckableState() clears the flags on the wallet software and the compatibility select, and why syncRequiredFields() clears the flag on hardware_works and seed_backup_verified. The rotation fields follow the same rule: the type is disabled until the key is marked for rotation, the path is disabled when the key has a saved check and is not marked, and the script refuses to send an empty path. The login page sidesteps it with two separate forms.
  • A hidden element has no box, so an anchor cannot jump to it. #history-detail is hidden until a check is opened, so revealDetail() in history.js unhides the section first and only then calls scrollIntoView(). Do not reorder those two lines.
  • A bare substring assertion over page markup also catches comments. Assert on the element, not on the phrase.
  • A display: grid panel ignores the hidden attribute. .menu-panel[hidden] { display: none; } is in the sheet for that reason, the same way .form-stack[hidden], .field-grid[hidden] and .form-section[hidden] already were.
  • The menu's outside-click listener has to allow the button's own click. The button's handler runs first and toggles the panel; the document listener then sees a target inside .menu and does nothing.
  • The current-page check compares a normalised path, and the current item is a span. currentPath() strips trailing slashes and keeps the root as /. The match is rendered as a <span class="menu-item menu-item-current"> with aria-current="page", never an <a>.
  • A sticky top bar needs a scroll container that is the viewport. .topbar is a flex child of body, and body sets no overflow, so position: sticky; top: 0 works — but it stops working the moment an ancestor gets overflow: hidden or auto. The menu panel is position: absolute inside the bar, so it travels with it.
  • The mismatch confirmation follows its warning. #address-path-confirmed is required only while #address-path-warning is visible, so the browser itself stops the submission.
  • The per-field secret reminder is gone; only the setup keeps one. A test asserts setup.html holds exactly one class="secret-warning" and that the check form and the settings hold none.
  • static/health-check.js must not use placeholder anywhere. A test asserts the string is absent, because a previous value is a real input value.
  • The key card's line order is pinned by two test modules. tests/test_key_status.py asserts card.append(top, name) → card.append(model) → card.append(path), and tests/test_health_check_form.py asserts the top line, that card.append(type) is gone, and that the model line is skipped by sameLabel(entry.model, entry.name).
  • The dashboard heading is dynamic and the static description is gone. tests/test_app.py asserts id="setup-shape" and the fallback text on the dashboard, while tests/test_pwa.py still asserts the manifest description. Do not move the one into the other.
  • The stylesheet test asserts on the shape of the theme, not on its wording. It checks the two bevel colour pairs, the dotted focus outline, the font stack, the absence of box-shadow and gradient, and that every border-radius in the sheet is 0.
  • The brand mark is an image, and the class name is pinned. IconTests in tests/test_pwa.py asserts that all six pages carry <img class="brand-mark" src="/icon.svg" alt="" aria-hidden="true">, that the sheet holds image-rendering: pixelated, and that place-items: center appears nowhere. It also pins min-height: 100vh, min-height: 100dvh, flex-direction: column and the exact string main { flex: 1 1 auto; }. Do not style the brand mark as a letter again.
  • The two SVG files must keep the same drawing. The package icon at the repository root is the source of truth, and test_the_served_icon_carries_the_same_drawing_as_the_package_icon compares them from <svg> onwards. A change to one drawing has to be made in both.
  • The two PNGs are real files, rasterized by the user, because the agent has no rasterizer. IconTests reads each file's IHDR and asserts 192×192 and 512×512.
  • The manifest lists the SVG first, and the two PNGs after it. test_the_manifest_lists_the_served_icon reads icons[0]. The PNG entries must carry the real size (192x192 and 512x512) and type: "image/png".
  • The favicon link sits next to the manifest link, and /login has no manifest. IconTests asserts the exact string <link rel="icon" type="image/svg+xml" href="/icon.svg"> on all six pages.
  • make x86 runs prettier --check startos, and the config is the SDK's own. .prettierrc is the single line "@start9labs/start-sdk/prettier.config.json", which is { trailingComma: "all", tabWidth: 2, semi: false, singleQuote: true } — everything else is prettier's default, so printWidth is 80 and quoteProps is "as-needed". Write it the way prettier would, or run npx prettier --config node_modules/@start9labs/start-sdk/prettier.config.json --write startos.
  • The page heading uses class="eyebrow" too. tests/test_health_check_form.py asserts the eyebrow lists of health-check.html and settings.html exactly, so a new eyebrow breaks it, and so does a removed one. The secret warnings deliberately use class="secret-warning", not an eyebrow.
  • A static help line that the script overwrites is dead text. #software-help in health-check.html is empty, because lockSoftwareFields() fills it before the user can read it. A test pins both halves: the empty element in the markup and softwareHelp.textContent in the script.
  • aria-labelledby may point at the eyebrow. login.html's first form has no h2; its aria-labelledby="login-form-title" points at the eyebrow <p> that carries the same id.
  • Do not assert on the quoting style of the JavaScript. static/health-check.js uses single quotes for constant selectors and template literals where a ${...} substitution is needed, and static/settings.js uses double quotes. Assert on the selector text and on quote-free substrings.
  • clean_record() in tests/test_alerts.py has a fixed provider list and carries the key on the security rows. Each ai_security_checks entry is a dict with provider, key_id and status: the entry whose provider is the key's own name carries the key number, and the others carry None. The alert texts are asserted there, and not in tests/test_app.py, which checks the code, the severity, the source, the actions and the key_id only.
  • sample_payload() / unavailable_payload() are duplicated in tests/test_validation.py, tests/test_database.py, tests/test_app.py, tests/test_maintenance.py and tests/test_secret_guard.py. They must send the key's own address_path, because the server compares it with the stored one.
  • Every test fixture uses one address path per key (ADDRESS_PATH, SECOND_ADDRESS_PATH, THIRD_ADDRESS_PATH). tests/test_app.py also has ROTATION_PATHS and the rotate() helper. With schema 13 a rotation retires the key row, so a test that rotates and then records a check has to use the new key id — rotate() hands it back under key_id.
  • validate_health_check() takes a third argument (the wallet software list), and validate_rotation() takes a fourth for the same reason.
  • tests/test_app.py has seven test classes. A new test must go in the class whose setup it needs: ServerTestCase, AppHandlerTests, RenamedSetupTests, SettingsApiTests, RotationApiTests, AcknowledgementApiTests, AuthApiTests and KeySetupApiTests. Every request helper sends the cookies and the CSRF header, and assert_status() is the one place an error status is expected.
  • The other server test modules share tests/server_session.py. test_export_filename.py, test_key_model.py, test_key_status.py, test_multisig_setup.py, test_pwa.py and test_secret_guard.py mix in SessionMixin, which signs in through the API; each class calls authenticate() at the end of its own setUp. A test that switches DATABASE_PATH to another file has to call authenticate() again, because the session secret lives in the database.
  • A login lockout test has to clear the counter. AuthApiTests shares one database per test, and the lockout lives in app_settings, so a test that trips it leaves the account locked for the tests that follow unless it resets the two rows or uses its own database file.
  • The export link and its note live on the history page, and four tests moved with them. tests/test_export_filename.py asserts the dashboard has no /api/export link; tests/test_app.py's test_the_dashboard_carries_no_navigation_link_of_its_own; test_the_history_page_offers_the_export_in_its_own_section in tests/test_health_check_form.py pins the Export section, the title bar, the note under the button and the absent .page-heading > .secondary-button rule; and test_the_history_links_to_the_detail_section pins both anchors.
  • #key-count still exists in setup.html, but not on the dashboard. The two are on different pages and were never the same element, so assert per file.
  • Section 19 records one key per submission, so a Check id in the export can cover up to three keys only for rows saved before section 19.
  • The database path default is repeated in sixteen places. database.py carries database_path="data/vardkey.sqlite3" as the default of every storage function (fifteen of them), and app.py carries it in get_database_path(). A missed one makes a single code path read or create the other file, silently.

Tooling — what the agent can and cannot do

  • The Gitea tools write a whole file at a time (gitea_update_file, auto-SHA, one file per call), and there is no preview step and no shell. There is no branch delete either.
  • The agent therefore cannot run the suite, make x86, tsc --noEmit or prettier --check. The user runs them, and the user is the one who rasterizes an image or deletes a branch by hand.
  • Keep the documents short. TODO.md and this note are read by every future model, and their size is context the next round cannot spend.

Product clarification

The key's configured name is its product label, and a cloud key's name is also a wallet-software choice. A hardware key name is never a wallet-software choice, and a cloud key records its own name as its software. A key is identified by its type and its address path, while the name is only a label that the user may change. The address path is unique per key, and a rotation replaces the key with a new one that must carry a new address path and may carry the other type. A rotation creates a new wallet, so the funds have to be moved from the old multisig setup to the new one before the old keys are discarded. An alert is closed by the user with the action that belongs to it, and a key with an open alert cannot be checked until the alert is closed. Since §32.8 point 5, a rotation also retires the old key row: the multisig slot number stays, the old checks keep the name and type they were recorded with, and the new row starts with an empty history.