Files
vardkey/ai-note.md
T

27 KiB
Raw Blame History

AI Agent Note — Helsesjekk

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

  • TODO.md (Norwegian) is the requirement specification: sections 1–18 are the spec, sections 19–22 a short appendix with the four decisions taken along the way, section 23 the configurable key setup, section 24 the required choices and the removal of BIP327/SLIP39, section 25 the address path identity, the key rotation, the alert acknowledgement, and the settings page, and section 26 the blocking rule, the unique address paths, the rotation with a new path, and the per-alert actions.
  • 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  • HANDOVER.md was deleted as outdated. Do not recreate it.

Status

  • Done and verified: sections 19–22, section 18 steps 1–8, and sections 23, 24 and 25.
  • Section 26 is implemented, not yet verified. All the code and the tests are written, README.md and this note are updated, and TODO.md still shows section 26 as open on purpose: the boxes are ticked once the user has confirmed that the suite is green on Linux and on the Windows copy.
  • Verified how: the automated suite and the user's manual browser testing, which is also where section 26 came from. The remaining manual tests are the ones in section 16 of TODO.md; the StartOS test there is still open.
  • Tests: the user runs them: python3 -m unittest discover -s tests. They run on a Windows copy at C:\Users\Driftsleiar\Downloads\helsesjekk, so they must pass there as well as on Linux.
  • Next work: run the suite, fix whatever it reports, then tick section 26 in TODO.md.
  • After that: section 18 steps 9–12 — the StartOS package, the persistent volume, backup/restore, and .s9pk. Covered by sections 13, 14, 16 and 17 of TODO.md. Do not start the StartOS part before the user says so.
  • Future work, deliberately not built: more than three keys, and another threshold than 2-of-3 (section 23).

Section 26 — implemented, waiting for the test run

Read section 26 of TODO.md for the full list. The parts that are easy to get wrong:

  • A key with an open alert cannot get a new check. POST /api/health-checks answers 409 for that key, and the check runs before the rest of the validation, so a blocked key answers 409 even when the payload would otherwise be a 400. The blocking applies to that key only, and the key is free again as soon as its alert is closed. The form marks the blocked options as disabled with a — blocked by an open alert suffix, and #blocked-keys-note names the alert, the action that closes it, and the keys that are still free.
  • Every alert must stay closable. _alert() raises ValueError when the action list is empty, so an alert without an action is a bug and not a state the user can reach. test_every_alert_rule_produces_at_least_one_action sweeps every rule and asserts that the covered codes are exactly ALERT_CODES.
  • The actions are fixed per code and source. ROTATE, REPLACE_HARDWARE, REMOVE_SOFTWARE, IGNORE and CONTACT_UPDATED replaced the generic ACKNOWLEDGE. See the table in section 26.4 for which code offers what: hardware_failure lost the generic acknowledgement because replace_hardware does the same job, and security_compromised has no ignore, whether the source is the key or the software.
  • rotate is not an acknowledging action. CLOSING_ACTIONS is the set the acknowledgement endpoint performs, and validate_acknowledgement accepts only those, so posting rotate is a 400. A rotation goes through POST /api/keys/<id>/rotate, which requires a new address path, so a key alert cannot be closed without replacing the key.
  • The address path must be unique across the three keys. Section 25 allowed the same path for two keys; section 26.2 does not. The comparison happens after strip and without regard to case. The setup answers 400, the settings 409, and the rotation 400, and the message names the key that already has the path. DuplicateAddressPath(ValidationError) is what keeps the 409 in the settings apart from the other validation errors, without matching on message text. No legacy handling is needed: the user deletes the data folder and configures the setup again.
  • A rotation now requires a new address path. address_path_unchanged is gone. The payload is replaced, keep_name, name and address_path: an empty value, the stored path, and a path another key already has are all refused with 400. The key number stays and the name may be kept or changed. Each rotation card has its own required New address path field, enabled when the key is selected, and the script refuses to send a rotation with an empty path.
  • A software alert names the software. software_incompatible and a software-sourced security_vulnerable use the software name as source and carry it in a new software field, so the settings link, the button and the removal point at the right entry. A record without a recorded software name keeps the stable source Software compatibility and a null software.
  • Removing a wallet software closes the alerts it raised. _close_removed_software() in app.py acknowledges the alerts whose software matches a removed name, through software_alert_targets(), and the stored reason is remove_software. The settings answer names the closed alerts in acknowledged. The dashboard offers the removal as a plain button when the failing name is in the list and the list holds more than one entry, and a link to the settings otherwise, because the list may never be empty.
  • The dashboard key card shows the open alert. It shows Needs action instead of its ordinary status, how many alerts are open, the action that closes the first one, and a link to the alert list.
  • The action wording lives in the frontend. ACTION_LABELS and ACTION_CONFIRMATIONS are duplicated in static/app.js and (labels only) in static/health-check.js, which matches the existing pattern of small helpers per file. The server sends action names only.
  • One CSS change plus one small block. .rotation-list { margin-top: 22px; } is the section 26.6 spacing fix; .status-alert, .key-alert-note, .key-alert-message and .key-alert-detail are the small style block for the new key-card message.

Section 25 — implemented and verified

Section 25 of TODO.md is the short summary; what follows are the parts that are easy to get wrong.

  • The address path is the identity of a key, together with its type. keys.address_path was added (a non-empty value in the setup, TEXT NOT NULL DEFAULT '' in the table), /setup asks for it, and the form records only Address path, pre-filled from key.address_path. The name is a label that the settings may change.
  • The server derives address_path_verified. It compares the recorded path with keys.address_path after stripping both: equal gives OK, different gives PROBLEM. A submitted address_path_verified is ignored (the user chose alternative A): the server always writes its own value, so a stale client that claims OK is overruled and an alert can never be lost. The comparison only happens when available is 1, so an unavailable key never reports a mismatch and never raises the mismatch alert.
  • A mismatch is treated like a key that could not be checked. address_path, the key note, the contact confirmation and the overall comment are stored; the hardware, the software and the security results stay empty, wallet_software is NULL, and software_compatibility_status is NOT_ASSESSED. Sending software or security results with a mismatched path is a 400, exactly as section 22 does for an unavailable key. software_version follows section 22 as well: it is read as text and stored.
  • The form warns before it submits. #address-path-warning appears when the recorded path differs, with the checkbox #address-path-confirmed. That checkbox is required only while the warning is visible, so a required control is never hidden. When the path does not match, the browser builds the minimal payload itself and the review summary marks the choice.
  • The wallet software list replaced the built-in Sparrow choice. validation.py no longer has SPARROW_SOFTWARE; the choices are software_names from the settings plus the configured cloud key names. POST /api/settings/wallet-software stores the whole list, which must hold at least one name, may not repeat a name, and may not contain a name that a cloud key already provides. A new database is seeded with Sparrow Wallet, so the user must add an alternative before the list can be emptied.
  • /settings is new and is in STATIC_FILES. It holds 01 / Keys, 02 / Wallet software and 03 / Key rotation. The keys section saves one key at a time (POST /api/settings/keys), the address path input is disabled when the key has checks, and has_checks from GET /api/settings tells the page which keys those are.
  • Rotation lives only in the settings. POST /api/keys/<id>/rotate updates the key row, closes the alerts its own checks raised, and changes no earlier check. The page can rotate several keys in one action, but it sends one request per key, because section 11 fixes the endpoint per key. The page states that rotating a key creates a new wallet, so the funds must be moved from the old multisig setup before the old keys are discarded. (Section 26.3 replaced address_path_unchanged with a required, unique, new path.)
  • Acknowledging an alert stores a fifth field. alert_acknowledgements holds code, source, reason, acknowledged_at and check_id (the user chose alternative B). An alert is hidden while alert.check_id <= acknowledgement.check_id, so a newer check, which always has a higher id, raises it again whatever date it carries. The date-comparison variant was rejected because a new check saved on the same day as the acknowledgement would stay hidden. (Section 26.4 stores the action name in reason.)
  • Every alert carries actions and key_id. Section 26.4 replaced the action set, added the software removal and added the enforcement of which action may close an alert.
  • The two dead alert rules are gone. cloud_provider_problem and address_verification_failed were removed, and the merged cloud wording of key_unavailable went with them, so one message covers every key.
  • Schema version 10. keys.address_path, the wallet_software table and the alert_acknowledgements table were added, and cloud_health_status and cloud_provider_operational were dropped from key_checks. There is no data migration for an older test database: the user deletes the data folder and configures the setup again. _key_checks_need_rebuild also rebuilds when it still sees the two cloud columns. Section 26 adds no columns or tables, so the schema stays at version 10.
  • The form label and the cloud version field. Software version (if used) became Software version, and the version field moved out of the grid that is hidden for a cloud key, so every key has it. The version is still reused from the last saved value for the same key and software.
  • A latent bug was fixed on the way. The grid with the wallet software and the compatibility result is hidden for a cloud key, and those two controls used to stay required, which blocks a submission with "An invalid form control is not focusable". syncUncheckableState now clears the flag for a cloud key as well.

Section 24 — implemented and verified

Read section 24 of TODO.md for the summary. 24.2 and 24.4 were delivered together on purpose, so the schema went to version 9 in one step.

  • 24.1: Hardware works is a required Yes/No with Select status as a non-answer first option, for an available hardware key only; the required flag is cleared when the key is unavailable. The contact confirmation is a required Confirmed current/Needs external update, also for an unavailable key. NOT_ASSESSED survives only as the derived software_compatibility_status of a key that could not be checked.
  • 24.2 and 24.4: the whole BIP327/SLIP39 function is gone (form, validation, alerts, history, CSV columns), and schema version 9 drops the tool_checks table with its rows.
  • 24.3: the form is renumbered to 01 Schedule, 02 Key, 03 Compatibility, 04 External research, 05 Contact confirmation, 06 Notes. 04 / Standards is gone.
  • 24.5: the BIP327/SLIP39 tests were removed, tests for the required choices and for the version 8 → 9 migration were added, and README.md was updated.

Section 23.2–23.5 — implemented and verified

  • The setup is stored once. POST /api/setup writes the three key rows; a second call is a 409, because the type and the address path are locked (section 23.2 and 25.1). GET /api/setup answers {"configured": …, "keys": […]}.
  • A new database shows the setup step. /setup is a page in STATIC_FILES, and static/app.js, static/health-check.js and static/settings.js send the browser there when the setup is missing.
  • An existing database is migrated, not replaced. list_keys() reads the stored rows. A database with 1 or 2 key rows raises RuntimeError and is never answered with the setup step.
  • The name is the product label. ai_security_checks.provider holds the configured key name and the selected wallet software, so validation.py builds the list from the key, and static/health-check.js does the same in expectedLabels().
  • The type decides, never the number. _key_check() and _compatibility_status() switch on key["type"], and the script does too.
  • The wallet software choices are the settings list plus every configured cloud key name. A hardware key name is never a choice. A cloud key records its own name, because the provider is the software for that key.
  • The setup is validated. validate_key_setup() rejects a wrong number of keys, a repeated or out-of-range key number, an empty or over-long name, a name that repeats another one without regard to case, an unknown type, an empty or over-long address path, and — since section 26.2 — a path that another key already has. Name and path uniqueness are enforced there and not by a UNIQUE constraint, so a legacy database with odd rows is refused by the API instead of being impossible to open.

Known leftovers

  • /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: removing them would shrink the endpoint, but nothing breaks while they stay.

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 (only NOT_ASSESSED, "" or null may accompany it).
  • Availability is required and offers only Yes/No. There is no "not assessed" value for it, and the server rejects a payload without it.
  • A cloud key collects only Availability. Cloud health, "provider operational" and software compatibility are all derived from it, and its own name is the wallet software. The two derived cloud columns and their alerts are gone (section 25.6).
  • A hardware key needs an explicit compatibility status. OK or PROBLEM; a missing value and NOT_ASSESSED are both a 400 (section 23.1).
  • Hardware works needs an explicit Yes or No for an available hardware key (section 24.1). A missing value, "" and NOT_ASSESSED are all a 400; only a key that could not be checked leaves it empty.
  • 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 (section 24.1).
  • NOT_ASSESSED is not offered or accepted anywhere else (section 24.1). No field but the derived software_compatibility_status accepts it.
  • Address path verified is never a form field again. The server derives it from the comparison (section 25.1). The stored column stays, so the history, the export and the alerts keep working.
  • The address path is locked after the first saved check. POST /api/settings/keys accepts the stored path again but answers 409 for a different one. A rotation is the way to give a key a new path (section 26.3), and it must be a new, unique one.
  • No two keys may share an address path, in the setup, in the settings or in a rotation (section 26.2).
  • /api/alerts covers every saved check, not only the newest, merges identical alerts by code plus source, and leaves out the acknowledged ones. A security alert's source is the provider, which is the key's configured name or the software name, and a software alert also carries that name in software.
  • An alert is closed by the user, not by a newer check. A newer check that raises nothing does not clear an older alert; the alert is closed by the action that belongs to it, and a key with an open alert cannot get a new check at all (section 26.1).
  • An acknowledgement stores the action name, not a free reason (section 26.4), and only the actions the endpoint performs are accepted.
  • The export is plain CSV only. XLSX is not built and the app does not promise Excel compatibility.
  • The prohibition lists in sections 1 and 13 of TODO.md stay unticked. They are standing prohibitions, to be re-confirmed before release.
  • Whether a date is overdue is decided in the browser, so /api/key-status stays independent of the current date.

Data model (schema version 10)

  • keys: key_number (1–3), name, type (hardware/cloud), address_path, 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, and the path must be unique across the three rows (section 26.2).
  • 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. wallet_software and contact_info_verified are free text.
  • key_checks: one row per submission — available, hardware_works, firmware_version, seed_backup_verified, address_path_verified, address_path, comments. The two cloud columns are gone.
  • ai_security_checks: provider (free text), status, comments, with UNIQUE (health_check_id, provider).
  • wallet_software: name, created_at. Stored whole, seeded with Sparrow Wallet when the list is empty, never emptied through the API.
  • alert_acknowledgements: code, source, check_id, reason, acknowledged_at. reason holds the action name that closed the alert (sections 26.4 and 26.7), so no schema change was needed.
  • tool_checks no longer exists. Version 9 dropped it with its rows (sections 24.2 and 24.4).
  • Startup migration rebuilds keys when it still has device, key_checks when it uses the old layout or still holds the two cloud columns, and health_checks and key_checks when their CHECK constraints still allow the old values. It then drops tool_checks if it is there.
  • 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 and SCHEMA_VERSION_9 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.

API

Pages: /, /setup, /settings, /health-check, /history. Endpoints: GET /health, 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 README.md.

Section 26 changed the rotation payload to replaced + keep_name + name + address_path, made POST /api/alerts/acknowledge take action, made POST /api/health-checks answer 409 for a blocked key, and added the uniqueness rule to three endpoints (400 in the setup and the rotation, 409 in the settings).

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 and /settings.js.
  • /api/setup, /api/settings, /api/alerts, /api/key-status and /api/previous-values are extra endpoints that are not in the minimum list in section 11. They are noted in section 11 but not in the specification itself.
  • 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.
  • 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 must be caught before the general ValidationError in _save_key_settings, or the 409 turns into a 400.
  • 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.

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 required control inside a display:none container still blocks submission. The browser refuses to submit and logs "An invalid form field is not focusable". Hiding a section is not enough: the required flags must be cleared, or the section must render something other than required fields. 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 — also for a cloud key, whose grid is hidden — and why syncRequiredFields() clears the flag on hardware_works and seed_backup_verified. The new rotation path field follows the same rule: it is disabled until the key is selected, and the script refuses to send an empty one.
  • 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 and a required control is never hidden.
  • The page heading uses class="eyebrow" too. A regex over health-check.html finds Annual security check before the numbered sections, and settings.html starts with Settings. A test on the section order must expect that first. The blocked-keys note adds no eyebrow of its own, so the order test is unchanged.
  • 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, e.g. [name="address_path"], and on quote-free substrings such as /api/settings and /settings?rotate=.
  • clean_record() in tests/test_alerts.py has a fixed provider list and carries key_id and wallet_software. A test that needs another provider must replace the entry itself, and the recorded software is Sparrow Wallet, which is what a software alert names.
  • sample_payload() / unavailable_payload() are duplicated in tests/test_validation.py, tests/test_database.py and tests/test_app.py. They build the labels from the key's name and the selected software, they always send contact_info_verified, and for a hardware key hardware_works. They must send the key's own address_path, because the server compares it with the stored one; tests/test_app.py and tests/test_database.py also have a mismatch_payload() for the minimal mismatching submission.
  • Every test fixture now uses one address path per key (ADDRESS_PATH, SECOND_ADDRESS_PATH, THIRD_ADDRESS_PATH), because the setup refuses duplicates. tests/test_app.py also has ROTATION_PATHS and the rotate() helper, which picks an unused path per key so a test can rotate several keys in one run.
  • validate_health_check() takes a third argument. The wallet software list from the settings must be passed in, or a hardware key that selects a settings name is refused. tests/test_validation.py and tests/test_database.py wrap it in a small validate() helper for that reason.
  • tests/test_app.py has six test classes. ServerTestCase starts the server, AppHandlerTests stores the default setup in setUp, RenamedSetupTests stores the renamed one, SettingsApiTests, RotationApiTests and AcknowledgementApiTests store the default one, and KeySetupApiTests leaves the database empty. A new test must go in the class whose setup it needs.
  • 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.

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. 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.