tasks/get examples under "Task Execution Errors" use "resultType": "task"; prose and schema require "complete"

[davidscotttufts]

What's wrong

Two example responses in Task Execution Errors show a tasks/get result with "resultType": "task":

• specification/2026-07-28/tasks.md:846, in "Example: Task with JSON-RPC execution error"
• specification/2026-07-28/tasks.md:870, in "Example: Tool call completed with tool error (isError: true)"

specification/draft/tasks.md has the same text at the same line numbers.

Both are tasks/get responses. The section says so (tasks.md:835: "The tasks/get response SHOULD include a statusMessage…"), and the text after the second example calls it "The tasks/get endpoint" (:889).

The normative text says the opposite:

• tasks.md:338: "The resultType field MUST be set to "complete" on this object as it is the standard result shape for the tasks/get request."
• tasks.md:102: "Servers MUST NOT set resultType to "task" on result types other than CreateTaskResult."
• schema/2026-07-28/schema.ts:224 and schema/draft/schema.ts:224: GetTaskResult has resultType: "complete";.

The other tasks/get examples in the same file already use "complete" (:577, :610, :655, :733, :762).

Why it matters

Implementers copy examples. A client that switches on resultType and treats "task" as "this is a CreateTaskResult, start polling" would handle a terminal tasks/get response as a new task handle. A server built from these examples would break the MUST at :338.

Suggested fix (docs only)

--- a/specification/2026-07-28/tasks.md
+++ b/specification/2026-07-28/tasks.md
@@ -843,7 +843,7 @@
   "jsonrpc": "2.0",
   "id": 4,
   "result": {
-    "resultType": "task",
+    "resultType": "complete",
     "taskId": "786512e2-9e0d-44bd-8f29-789f820fe840",
     "status": "failed",
@@ -867,7 +867,7 @@
   "jsonrpc": "2.0",
   "id": 5,
   "result": {
-    "resultType": "task",
+    "resultType": "complete",
     "taskId": "786512e2-9e0d-44bd-8f29-789f820fe840",
     "status": "completed",

Apply the same two changes to specification/draft/tasks.md (same line numbers). Nothing normative changes.

[davidscotttufts]

Disclosure per the MCP AI_POLICY.md: this issue was drafted primarily with Claude Code; I reviewed it and checked the cited lines against the spec and schema. The fix is in #25.