Three breakages on the prior CI run:
1. PHPUnit 12 didn't propagate `@depends` data — chained tests got
ArgumentCountError because they expected `array $data` but PHPUnit
passed nothing. Convert all annotations to `#[Depends]` attributes,
matching the modern style already used by the Migrations and
VectorsDB suites.
2. `InsightsCustomConsoleTest` extends ProjectConsole, which doesn't
set up a project API key. The trait's `serverHeaders()` hardcodes
`x-appwrite-key`, so every Console test 401'd. Drop the Console
class entirely — the manager Create endpoint is KEY-only by
design, the Server class already exercises every code path, and a
Console-side variant adds no real coverage.
3. `testCreateRequiresManagerScope` called `getNewKey()`, which lives
on `ProjectCustom`. PHPStan flagged the call as undefined when the
trait was analyzed against the (no-longer-existing) Console class.
Move the test into `InsightsCustomServerTest.php` directly so it's
only ever resolved against `ProjectCustom`.
Plus PHP 8.4 + match-exhaustive cleanups PHPStan caught while I was
in there:
- `?array $headers = null` instead of `array $headers = null` on every
helper (PHP 8.4 deprecates implicit-nullable params).
- `?string $insightId = null` on `sampleInsight()`.
- `parentResourceType` match collapsed to `tablesDB => 'tables';
databases/documentsDB/vectorsDB => 'collections'`. The earlier
`$type` match already throws on unknown engine, so the fall-through
default was unreachable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Insights are children of reports — make the URL hierarchy reflect that.
Endpoints:
- POST /v1/manager/reports/:reportId/insights (manager Create)
- GET /v1/reports/:reportId/insights (List)
- GET /v1/reports/:reportId/insights/:insightId (Get)
- PATCH /v1/reports/:reportId/insights/:insightId (Update)
- DELETE /v1/reports/:reportId/insights/:insightId (Delete)
`reportId` moves from optional body field to required path param. All
endpoints fetch the report first (404 REPORT_NOT_FOUND if missing or
in another project), then verify the insight's `reportInternalId`
matches before doing anything else.
Side effects:
- Event names nested: `reports.[reportId].insights.[insightId].create`
etc. Top-level `insights.*` event tree removed from events.php.
- Realtime channel parser handles the nested form: a `reports.{rid}`
event lights up `reports`, `reports.{rid}` channels; a nested
`reports.{rid}.insights.{iid}` event also lights up
`reports.{rid}.insights` and `reports.{rid}.insights.{iid}`.
- Audit resource paths nested similarly:
`report/{request.reportId}/insight/{response.$id}`.
- listInsights query validator drops `reportId` from
ALLOWED_ATTRIBUTES — it's path-scoped now, not a query filter.
Tests:
- E2E helpers `createInsight`/`getInsight`/`listInsights`/
`updateInsight`/`deleteInsight` all take `reportId` as the first
argument.
- New `createFixtureReport()` helper for standalone validation tests
that need a parent.
- Dropped `testCreateWithoutReport` — reportId is mandatory now.
- `testCreateRejectsUnknownReport` now exercises the path-level 404
rather than a body-level check.
- `testGet` and `testUpdateMissing` exercise the
wrong-reportId-but-valid-insightId case (returns
`report_not_found`).
- `testList` asserts every result carries the path's reportId, plus a
404 case for a nonexistent parent.
- `testCreateForEachEngine` and the standalone create-rejection tests
inline-create their own fixture report and clean up after.
- `testListSurvivesEmptyDatabase` renamed to
`testListSurvivesEmptyReport` and uses a fresh fixture report.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User Update only accepts severity + status now — analyzer-controlled
fields like summary flow through the manager Create endpoint. The
previous testUpdate sent `summary: 'Updated summary.'` and asserted
it changed; Utopia silently dropped the unknown param so the
assertion would have failed.
Trim the call to severity-only and verify the analyzer fields
(including summary) are preserved on the response.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`key` was a leftover from when CTAs were embedded JSON — there's no
remaining reason to require analyzers to invent a within-insight
identifier. The execution layer is gone (no `cta.key` event format),
insights are immutable from the user side (analyzers re-ingest by
delete + recreate, so idempotent matching never happens), and `label`
already covers human-facing identification. The console can group/sort
CTAs by `service`+`method` if needed.
- Schema: drop `key` attribute and the UNIQUE
`(insightInternalId, key)` index from insightCTAs. Required fields
are now `label`, `service`, `method` (+ optional `params`).
- Validator no longer requires `key`. Drop the dup-key normalization
loop in the manager Create endpoint — there's no semantic
uniqueness to enforce.
- Response model: `InsightCTA` keeps `$id` + standard headers,
`insightId` backref, and the four functional fields.
- E2E: drop sampleCTA's `$key` parameter, drop the
testCreateRejectsDuplicateCTAIds test entirely, rename empty-fields
test to testCreateRejectsCTAWithEmptyLabel and update the missing-
fields tests to drop `key` from their payloads.
- Unit tests rewritten to drop `key`.
- Comment on the `insights.ctas` virtual attribute updated to
reference the renamed `insightCTAs` collection.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Disambiguate the platform-level collection name. Field/request-param
remains `ctas` (the embedded array on the insight response).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Embedding CTAs as a 16384-byte JSON blob on `insights` was the wrong
shape — they're real documents with their own lifecycle. Move them out.
Schema:
- New platform `ctas` collection. Each row carries `projectInternalId`,
`projectId`, `insightInternalId`, `insightId` (backref), plus the
CTA fields: `key`, `label`, `service`, `method`, `params`.
- Indexes: `(projectInternalId, insightInternalId)` for the subquery
lookup and a UNIQUE `(insightInternalId, key)` so the per-insight
uniqueness invariant lives at the DB layer (not just in PHP).
- The `ctas` field on `insights` becomes a virtual attribute backed by
a new `subQueryInsightCTAs` filter that joins child docs at read
time. Consumers still get CTAs embedded on the insight response —
one round-trip from their perspective.
- The CTA descriptor's within-insight identifier renamed `id` → `key`
(clashed with the document `$id`). Validator updated.
Endpoints:
- Manager Create now persists CTAs as separate `ctas` documents after
the parent insight, then re-fetches the insight so the response
carries the freshly-joined CTA list.
- User Update trimmed to user-controlled state only (`severity`,
`status`). `title`, `summary`, `payload`, `ctas`, and `analyzedAt`
are analyzer-controlled — analyzers re-ingest by deleting and
POSTing again to the manager endpoint.
- Insight Delete cascades to CTAs.
- Report Delete cascades through Insights → CTAs.
Response model:
- InsightCTA gains the standard document headers (`$id`,
`$createdAt`, `$updatedAt`) and an `insightId` backref. The
caller-supplied identifier is now `key`.
Tests:
- E2E sampleCTA factory uses `key` everywhere; testCreate asserts the
freshly-created CTA carries `$id`, `$createdAt`, `insightId`, and
the right shape.
- Dropped the testUpdate*CTA* tests — user Update no longer accepts
CTAs. testDismissViaUpdate now depends on testUpdate directly.
- Unit tests rewritten to validate `key` instead of `id`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Eldad's review comment: insights about nested resources need a pointer
to the containing parent (the file-in-bucket pattern). Add three
optional fields:
- parentResourceType (plural noun, e.g. `tables`, `collections`)
- parentResourceId
- parentResourceInternalId
so an insight whose `resourceType=indexes` / `resourceId=_idx_status`
can also carry `parentResourceType=tables` / `parentResourceId=orders`
to identify the table that owns the index. All three are nullable for
top-level resources (e.g. a project-wide audit finding).
Schema, response model, manager Create endpoint, and the listInsights
query validator (parent fields are filterable). New compound index
`_key_project_parent_resource(projectInternalId, parentResourceType,
parentResourceId, $sequence)` to support the parent lookup pattern
the console will use ("show all insights for table X").
E2E factory generates a parent by default (engine-aware:
tables for tablesDB, collections for the others). New
testCreateWithoutParentResource exercises the top-level case;
testList gains a parent-resource filter assertion; testUpdate's
preserved-fields check picks up the new attributes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Insights are produced by internal Appwrite services (edge, executor,
background analyzers) — never by user clients. Move the ingestion
endpoint accordingly.
- Move Http/Insights/Create.php → Http/Manager/Insights/Create.php.
- Path: /v1/insights → /v1/manager/insights. SDK Method marked
`hide: true` and namespaced under `manager` so generated SDKs don't
expose it. Auth narrowed from [ADMIN, KEY] to [KEY] only.
- New scope `insights.manager`. Not granted by any user role
(app/config/roles.php) — Cloud/edge teams configure their internal
key issuance to grant it. `insights.write` description trimmed to
the user-facing surface (update/dismiss/delete) since create is now
manager-only.
- Reports, ListInsights, GetInsight, UpdateInsight, DeleteInsight
remain at /v1/insights/*. Existing scopes unchanged.
- Reports `categories` switched from JSON-encoded string to a native
array<string> column (size 64 per entry, up to 32 entries via the
endpoint validator). MySQL JSON-array indexes are weak and we never
query individual entries — read+rewrite only.
- E2E test API key in tests/e2e/Scopes/ProjectCustom.php gains
insights.read/write/manager + reports.read/write so the manager
endpoint is reachable from the test harness.
- E2E InsightsBase.createInsight() helper now POSTs /manager/insights.
- New testCreateRequiresManagerScope verifies a key with
insights.read/write but no insights.manager is rejected with 401.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
greptile flagged missing lengths but the codebase has two conventions:
- all-string composite: [LENGTH_KEY, LENGTH_KEY] (_key_unique, _key_provider_providerUid)
- mixed/non-string composite: [] (e.g. ('enabled', 'type'), ('region', 'accessedAt'),
('targetInternalId', 'topicInternalId'), ('period', 'time'), ('metric', 'period', 'time'))
(teamInternalId, confirm) is string+boolean; closest match is
('enabled', 'type') which uses lengths => [] and orders => [].
Aligning to that established pattern instead of inventing
[LENGTH_KEY, 0].
All four new indexes left lengths/orders as empty arrays; greptile
flagged the inconsistency vs every existing string-attribute index in
the file (e.g. _key_team uses [LENGTH_KEY], _key_unique uses
[LENGTH_KEY, LENGTH_KEY]).
- memberships._key_team_confirm: [LENGTH_KEY, 0] for (string, boolean)
+ [ORDER_ASC, ORDER_ASC]
- projects._key_teamInternalId: [LENGTH_KEY] + [ORDER_ASC]
- platforms._key_project_id: [LENGTH_KEY] + [ORDER_ASC]
- webhooks._key_project_id: [LENGTH_KEY] + [ORDER_ASC]
- InsightCTA model now exposes `service` (SDK namespace) and `method`
(function name) instead of a single `action` string. Drops the
string-splitting burden on every consumer and lets the console reach
the right SDK method directly.
- Validator requires both `service` and `method` non-empty; same
16-entry max still enforced.
- Endpoint normalization (Create + Update) splits the new shape into
the persisted CTA descriptor.
- Constants split: INSIGHT_CTA_SERVICE_* (databases / tablesDB /
documentsDB / vectorsDB) and INSIGHT_CTA_METHOD_* (createIndex).
- Insight model + InsightCTA model docs updated with the new field
semantics and per-engine examples.
- E2E factory `sampleCTA($id, $engine)` emits the correct service and
engine-appropriate params keys (tableId/columns for tablesDB;
collectionId/attributes everywhere else). Engine matrix asserts
`service` and `method` independently.
- Added e2e + unit coverage for the new failure modes (missing
service, missing method, empty service, empty method).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- memberships: _key_team_confirm on (teamInternalId, confirm) for team-membership confirm-state queries
- projects: _key_teamInternalId on teamInternalId for team-scoped project lookups
- platforms: _key_project_id on projectId for user-facing-id lookups
- webhooks: _key_project_id on projectId for user-facing-id lookups
Re-applies the indexes from the stale PR #9629 (1.7.x base, conflicting)
onto a fresh 1.9.x branch. None of these are in 1.9.x today; existing
similar indexes target projectInternalId / teamId rather than the
user-facing projectId / teamInternalId queries this addresses.
- Restructure InsightsBase trait with explicit helper methods
(createInsight/getInsight/listInsights/updateInsight/deleteInsight,
createReport/getReport/listReports/updateReport/deleteReport, plus
sampleInsight/sampleCTA factories) — same shape ProxyBase uses.
- Add coverage for: report CRUD + duplicate-id rejection, invalid type
rejection, list filtering by all allowed attributes, cursor
pagination + missing-cursor, update preserving untouched fields,
CTA validation edge cases (duplicate ids, empty fields, count > 16),
dismissal round-trip + status filter, report cascade delete,
unauthorized access (no server key), empty-result list.
- Engine-specific insight types (tablesDBIndex, documentsDBIndex,
vectorsDBIndex, plus the legacy databaseIndex) so the CTA's `action`
can map to the matching public API: databases.createIndex,
tablesDB.createIndex, documentsDB.createIndex,
vectorsDB.createIndex. dataProvider drives the engine matrix and
asserts the right action lands in the persisted CTA. Constants for
each action name live in app/init/constants.php.
- InsightCTA model docs spell out which action belongs to which engine
and that the params keys differ between APIs (tableId/columns for
tablesDB vs collectionId/attributes for the legacy / DocumentsDB /
VectorsDB APIs).
- Insight model `type` description now lists every engine variant.
- CTAsTest gains coverage for object-shaped params, empty-action and
empty-label rejection, and the default 16-entry cap.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Address review feedback on PR #12194:
- Pivot CTAs to pure descriptors (id/label/action/params). Drop the
server-side execution layer: Action interface, registry, the
databases.indexes.create CTA action, the params validator, the
/v1/insights/:id/ctas/:id/executions endpoint, the InsightCTAExecution
model, the INSIGHT_CTA_* errors, and the corresponding events. The
console invokes the existing public API directly with the descriptor's
action + params.
- Restore Databases\Indexes\Action.php to its pre-CTA shape and inline
the index-create body back into Create.php (the createIndex helper
was added solely for CTA reuse).
- Move insights collection from project DB to platform DB and add a
parent reports collection alongside it. Insights carry projectId /
projectInternalId for tenant scoping and an optional reportId for
grouping. List endpoints filter by projectInternalId; Get/Update/
Delete also enforce project ownership before touching the document.
- New Reports module with full CRUD (Create/Get/XList/Update/Delete),
Report response model, Reports query validator, REPORT_NOT_FOUND /
REPORT_ALREADY_EXISTS errors, reports.read / reports.write scopes,
and reports.* event tree. Delete cascades to child insights.
- Update.php now mutates the loaded document via setAttribute (instead
of passing a partial new Document), reuses CTAsValidator (instead of
the looser ArrayList<JSON> + isset check), and rejects duplicate CTA
ids.
- Create.php enforces unique CTA ids during normalization.
- CTAsValidator gained a configurable maxCount (default 16) so the
Create path matches the Update path and the DB column size, and
oversized payloads return a clean 400.
- Validator\Queries\Insights adds status and reportId to
ALLOWED_ATTRIBUTES so dismissal / report workflows are filterable.
- Realtime channel parser guards $parts[1] for both insights and
reports event names.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Each platform spec previously declared only the auth headers referenced by
endpoints in that platform's surface. The unified web SDK now exposes a
single Client class with cross-platform auth factories (fromSession,
fromCookie, fromDevKey, etc.), which means an SDK generator targeting one
platform still needs setter generation for headers that other platforms
expose. The web SDK currently papers over this with a webClientHeaders
augmentation in the SDK generator.
Add the missing securityDefinitions entries directly so every platform
spec carries the union, removing the need for SDK-side augmentation:
- client: + Cookie (for SSR cookie forwarding)
- server: + Cookie, DevKey
- console: + Session, DevKey
Per-endpoint security requirements continue to be filtered per platform
as before; only the securityDefinitions registry expands to declare
schemes that exist API-wide. SDKs that iterate the registry to emit
auth setters now produce the union without per-SDK fallback code.
copilot-swe-agent[bot]andabnegate
Jake Barnby
Matej Bačo
ArnabChatterjee20k
Damodar Lohani
Damodar Lohani
Matej Bačo
Chirag Aggarwal