test(blueprint): gate required graph output fields on the native surface

[Fl0p]

Context

Upstream reproduced get_node_details failing on the native transport with
OUTPUT_SCHEMA_VIOLATION: Missing required parameter 'nodeId' while A/B-testing the
posX/posY fix (ChiR24#592, merged 2026-08-10), together with create_reroute_node
returning an error receipt for a Knot it had just created.

Both are already fixed in dev — no code fix is proposed here:

defect	fix	landed
get_node_details omits nodeId	Result->SetStringField(TEXT("nodeId"), NodeId) in ...BlueprintGraphHandlersDetails.cpp	f4b9c66a, 2026-08-23
create_reroute_node omits nodeGuid	Result->SetStringField(TEXT("nodeGuid"), ...) in ...BlueprintGraphHandlersNodeMutations.cpp	52b5b8d2, 2026-09-04

What is missing is a gate. The same mistake landed twice in the same domain within a
month, and both times it broke the readback step of an operation that had actually
succeeded: the native execute seam projects a handler Result onto the declared output
contract and then validates it, so one unpublished required field turns a working call
into an error and the caller never receives the handle.

What this adds

tests/unit/plugin/blueprint_graph_output_field_contracts.test.ts, driven by the
generated canonical registry plus the GraphSubActions list in the domain registration
(no hand-maintained list of actions):

1. Domain-wide — every required output field of every graph sub-action is published by
   some BlueprintGraph handler. Sources are collected recursively, because
   list_node_types lives in Context/ and a root-only scan does not see it.
2. Handler-scoped — for the readbacks that build their own Result
   (get_node_details, create_reroute_node, get_graph_details, get_pin_details,
   list_node_types), the field must be published inside the function that claims the
   sub-action. create_node is deliberately excluded and documented: it finalizes through
   the shared node-creation path, so only tier 1 can speak for it.
3. Contract-level — the projected payloads are validated against their own output
   schema via the existing schema-subset mirror, so the omission fails as
   missing-required with the exact message the gateway reports
   (Missing required parameter 'nodeId').

Verification

Host is Raspberry Pi / ARM64 with no Unreal Engine, so nothing here was editor-verified —
but this change is tests-only and the gate itself was proven to bite:

• npx vitest run tests/unit/plugin/blueprint_graph_output_field_contracts.test.ts → 8 passed.
• With the two one-line fixes temporarily reverted in the working tree, the handler-scoped
  test fails and names both:

  + "blueprint.get_node_details.nodeId missing from McpAutomationBridge_BlueprintGraphHandlersDetails.cpp",
  + "blueprint.create_reroute_node.nodeGuid missing from McpAutomationBridge_BlueprintGraphHandlersNodeMutations.cpp",
  

• npx eslint on the new file → clean.
• npx vitest run tests/unit/plugin/ → 809 passed, 3 failed. All 3 failures pre-exist on
  clean dev (80430fd2) — verified by re-running them with this file removed:
  source_structure_contracts (McpAutomationBridgeHelpersProjectPaths.h is 267 pure lines
  vs the 250 gate) and the two native_discovery_* parity contracts under gateway/.

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 1c647ac8-d258-4962-b696-9c136832e5de

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Fl0p]

Closing: the defect class this gate memorializes no longer exists on dev.

Rebased the branch onto dev (60ad47b8) — clean, no conflicts — and ran the suite. 7 of 8 cases fail, and not because a handler regressed. The capability registry was consolidated in the meantime:

before	now
blueprint.get_node_details	legacy action of blueprint.inspect_graph (5 actions per record)
blueprint.create_reroute_node	legacy action of blueprint.edit_graph (9 actions per record)

Both consolidated records declare output.required = ['success']. nodeId and nodeGuid are now ordinary declared properties, so OUTPUT_SCHEMA_VIOLATION: Missing required parameter 'nodeId' — the failure this gate exists to prevent — is structurally unreachable on the graph surface.

I checked whether a quieter failure replaced it, since the projection keeps only declared properties. It did not: McpNativeGatewayOutputProjection.cpp:41-83 now folds every undeclared root field into a declared details object, and all three graph records (edit_graph, inspect_graph, delete_node) declare details. The 11 root fields the graph handlers publish outside the contract — x, y, nodeComment, nodeCount, conversionNodeId and the rest — reach the caller one level down rather than being dropped.

So both defects this branch guards were resolved by a change in the shape of the contract, not by the two one-line fixes it cites. A gate pinned to per-action capability ids cannot survive that, and rewriting it against the consolidated records would be a different test with a different premise.

Worth recording for later: the class is not extinct registry-wide. 34 records still declare a required output field beyond success/message, and one is unmet today — manage_tools.list_categories requires totalCategories, but FMcpDynamicToolManager::ListCategories() (McpDynamicToolManagerQueries.cpp:29) publishes only success and categories, while its neighbour ListTools() publishes totalTools. It surfaces as a silently absent required field rather than an error, because the local dispatch path (McpNativeTransportDynamicTools.cpp:41-52) answers directly and skips output projection and validation. That is its own fix plus a registry-wide gate, not this branch.