fix(server): watcher device-pin authz + table-schema drift test runs in CI#1062
Merged
Conversation
A watcher pinned to a device worker runs its agent CLI on the device owner's machine. manage_watchers create/update stored device_worker_id without verifying the caller may target that device, so a member-write actor could pin a watcher to another user's device (privilege escalation). Add a pure decision helper (evaluateDeviceWorkerAccess) plus a DB-backed assertion (assertDeviceWorkerAccess) mirroring watcher-execution-config: the caller must own the device, or be org owner/admin over a device attached to their org; system/internal callers are exempt. Wired into handleCreate and handleUpdate; rejects with ToolUserError (403). Unit-tests the role x ownership matrix (owner/admin/member/system x owned/foreign/missing) and adds an end-to-end gate assertion in watchers-crud (foreign-org pin rejected on create + update).
…efinitions.api_type The drift describe was gated on `describe.skipIf(!process.env.DATABASE_URL)` evaluated at module load. In the embedded-PG path global-setup sets DATABASE_URL in the setup process, but whether a forked vitest worker observes it at module-load time is timing-dependent, so the block could silently self-skip and the gate was toothless. Gate instead on the resolved URL global-setup now `provide()`s (vitest's transport-level channel that always reaches forks), read via `inject()` inside the test, skipping from the test body when no backend exists. Also add connector_definitions.api_type (a non-secret 'api'|'mcp'|'openapi' discriminator) to QUERYABLE_SCHEMA — it is a real DB column the drift test correctly flagged as missing; the large *_config blobs stay intentionally omitted.
📝 WalkthroughWalkthroughThis PR implements device-worker authorization for watcher pinning to prevent unauthorized cross-user/cross-organization device access, fixes a CI gap in schema drift testing by propagating database configuration to forked workers, and adds the missing ChangesDevice Worker Access Control and Test Fixes
Sequence DiagramsequenceDiagram
participant Handler as manage_watchers<br/>(create/update)
participant Assertion as assertDeviceWorkerAccess
participant DB as device_workers<br/>table
participant Decision as evaluateDeviceWorkerAccess
participant Error as ToolUserError
Handler->>Assertion: assertDeviceWorkerAccess(deviceWorkerId, caller)
alt deviceWorkerId is null/undefined
Assertion-->>Handler: return (skip validation)
else deviceWorkerId provided
Assertion->>DB: SELECT * FROM device_workers WHERE id = ?
alt device not found
Assertion->>Decision: evaluateDeviceWorkerAccess(null, caller)
Decision-->>Assertion: rejection reason
Assertion->>Error: throw 403 Forbidden
Error-->>Handler: authorization error
else device found
Assertion->>Decision: evaluateDeviceWorkerAccess(device, caller)
alt caller denied
Decision-->>Assertion: rejection reason
Assertion->>Error: throw 403 Forbidden
Error-->>Handler: authorization error
else caller allowed
Assertion-->>Handler: return (validation passed)
end
end
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Warning There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure. 🔧 ESLint
ESLint skipped: no ESLint configuration detected in root package.json. To enable, add Comment |
|
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@packages/server/src/tools/admin/watcher-device-access.ts`:
- Around line 82-90: The code must validate/normalize deviceWorkerId before
attempting the UUID cast in the SQL query: ensure deviceWorkerId (and the
computed trimmed) are checked for being a valid UUID format and not just
whitespace; if invalid, either throw ToolUserError(400) or return a normalized
null/UUID for the callers (handleCreate/handleUpdate) to persist. Specifically,
add validation logic around the deviceWorkerId/trimmed variables used before the
sql`... WHERE id = ${trimmed}::uuid` so non-UUID strings or pure-whitespace are
rejected or normalized, and ensure handleCreate and handleUpdate use that
normalized value instead of persisting the original invalid string.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: 055613a5-e932-48ad-9022-1ee0a38b6dae
📒 Files selected for processing (7)
packages/server/src/__tests__/integration/watchers/watchers-crud.test.tspackages/server/src/__tests__/setup/global-setup.tspackages/server/src/__tests__/unit/watcher-device-access.test.tspackages/server/src/tools/admin/manage_watchers.tspackages/server/src/tools/admin/watcher-device-access.tspackages/server/src/utils/__tests__/table-schema.test.tspackages/server/src/utils/table-schema.ts
Comment on lines
+82
to
+90
| if (deviceWorkerId === undefined || deviceWorkerId === null) return; | ||
| const trimmed = deviceWorkerId.trim(); | ||
| if (!trimmed) return; | ||
|
|
||
| const rows = (await sql` | ||
| SELECT id, user_id, organization_id | ||
| FROM device_workers | ||
| WHERE id = ${trimmed}::uuid | ||
| LIMIT 1 |
There was a problem hiding this comment.
Reject or normalize malformed device_worker_id before the uuid cast.
Whitespace currently skips the lookup here, but handleCreate/handleUpdate still persist the original value, so " " falls through to the later uuid write path and 500s. Any other non-UUID string will also 500 at WHERE id = ${trimmed}::uuid. Please validate this input up front and either raise a ToolUserError(400) or return a normalized UUID/null that the callers persist.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@packages/server/src/tools/admin/watcher-device-access.ts` around lines 82 -
90, The code must validate/normalize deviceWorkerId before attempting the UUID
cast in the SQL query: ensure deviceWorkerId (and the computed trimmed) are
checked for being a valid UUID format and not just whitespace; if invalid,
either throw ToolUserError(400) or return a normalized null/UUID for the callers
(handleCreate/handleUpdate) to persist. Specifically, add validation logic
around the deviceWorkerId/trimmed variables used before the sql`... WHERE id =
${trimmed}::uuid` so non-UUID strings or pure-whitespace are rejected or
normalized, and ensure handleCreate and handleUpdate use that normalized value
instead of persisting the original invalid string.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two follow-ups from the PR #1058 multi-aspect review (issues #1060, #1061). Both are server-only; no owletto/submodule changes.
#1060 — Validate
device_worker_idownership inmanage_watchersA watcher pinned to a device worker runs its agent CLI on the device owner's machine.
manage_watcherscreate/update storeddevice_worker_idwithout verifying the caller could target that device, so a member-write actor could pin a watcher to another user's device — privilege escalation (unattended agent execution on someone else's machine).Fix: new
watcher-device-access.tsmirroring the shape ofwatcher-execution-config.ts:evaluateDeviceWorkerAccess(device, caller)— pure decision function (no DB). Allows the device owner (device.user_id === caller.userId), or an org owner/admin over a device attached to their org (device.organization_id === caller.organizationId). System/internal callers (isAuthenticated && userId===null && memberRole===null) are exempt, matchingassertValidExecutionConfig.assertDeviceWorkerAccess(sql, deviceWorkerId, caller)— DB-backed wrapper;undefined(unchanged) /null/empty (clear) pass without a lookup; rejection throwsToolUserError(403).Wired into
handleCreateandhandleUpdateright afterassertValidExecutionConfig.handleCreateFromVersiondoes not setdevice_worker_id, so it needs no gate.#1061 — Drift test dormant in CI +
connector_definitions.api_typemissing(a) The drift
describewas gated ondescribe.skipIf(!process.env.DATABASE_URL)evaluated at module load — timing-dependent under forked vitest workers, so it could silently self-skip and the gate was toothless. Fixed by havingglobal-setupprovide('databaseUrl', …)(vitest's transport-level cross-process channel) and gating the test oninject('databaseUrl')read inside the test body, skipping only when there is genuinely no backend (SKIP_TEST_DB_SETUP=1).(b)
connector_definitions.api_typeis a real DB column (text DEFAULT 'api', a non-secret'api'|'mcp'|'openapi'discriminator) missing from bothQUERYABLE_SCHEMAandINTENTIONALLY_OMITTED, so the drift test fails against a real DB. Added it to the cols list (parity with the other scalarconnector_definitionsmetadata columns); the large*_configJSONB blobs stay intentionally omitted.Validation (Node 22)
make build-packages+cd packages/server && bunx tsc --noEmit— clean (exit 0).bun test watcher-device-access.test.ts): 9 pass (owner/admin/member/system × owned/foreign/missing). Integration (watchers-crud.test.ts, embedded PG18): 12 pass incl. 4 new device-gate cases. Teeth proven: removingassertDeviceWorkerAccessmakes the 3 rejection cases fail (a foreign-org pin silently succeeds) → restored → green.createdb wf_test→dbmate up→ vitest with thatDATABASE_URL: 14 pass, drift case shows✓not↓); and under the embedded PG18 backend (14 pass). Teeth proven: removingapi_typemakes it fail against the real DB → restored → green.dropdb wf_testafter.make review BASE=origin/main: typecheck=0, unit=0, integration=0; pi verdict bug_free 88, 0 bugs, 0 blockers, tests_adequate=true.Closes #1060
Closes #1061
Summary by CodeRabbit
New Features
Tests
Chores