fix(context): name emit_artifact in the schema instruction (#416)

The pinned schemaInstruction said only "Respond with a single JSON object
matching this schema", contradicting every role prompt that says to emit via
the emit_artifact tool. Two stated channels for one artifact, and the pinned
one won.

ResponseFormat.Json is built under exactly the condition that offers the tool
(an llmEmitted slot — see emitArtifactTool), so the tool is always available
wherever this instruction renders, and the kernel prefers it: tool-calling
models are more reliable at it and it sidesteps llama.cpp's grammar+tools
incompatibility. The instruction now names it, and keeps raw JSON as what to do
when told to stop calling tools — which is what the tools-less final pass
demands and what the executor already accepts either way
(llmArtifactOverride ?: response.text).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

core/kernel/src/main/kotlin/com/correx/core/kernel/orchestration/SessionOrchestratorContext.kt

@@ -52,11 +52,21 @@ internal fun SessionOrchestrator.evictArtifactContentCache(sessionId: SessionId)
 internal suspend fun SessionOrchestrator.buildSchemaEntries(
     responseFormat: ResponseFormat,
     stageId: StageId,
+    artifactName: String,
 ): List<ContextEntry> {
     if (responseFormat !is ResponseFormat.Json) return emptyList()
     val compactSchema = Json.encodeToString(JsonSchema.serializer(), responseFormat.schema)
-    val instruction = "Respond with a single JSON object matching this schema. " +
+    // #416: names emit_artifact as the channel. ResponseFormat.Json is emitted under exactly the
-            "Do not include markdown, code fences, or commentary outside the JSON. " +
+    // condition that offers the tool (an llmEmitted slot — see emitArtifactTool), so the tool is
+    // always available here, and the kernel prefers it: tool-calling models are more reliable at it
+    // and it sidesteps llama.cpp's grammar+tools incompatibility. The old text said only "respond
+    // with a single JSON object", contradicting every role prompt that says to call emit_artifact.
+    // Raw JSON stays valid as the second sentence describes, since the executor accepts either
+    // (llmArtifactOverride ?: response.text) and the tools-less final pass explicitly demands it.
+    val instruction = "Produce the '$artifactName' artifact by calling the $EMIT_ARTIFACT_TOOL tool " +
+            "with its fields filled in, matching this schema. If you are told to stop calling tools, " +
+            "output the same object as a single JSON object in your final message instead. Either way, " +
+            "no markdown, no code fences, and no commentary outside the JSON. " +
             "Schema: $compactSchema"
     return listOf(
         ContextEntry(

core/kernel/src/main/kotlin/com/correx/core/kernel/orchestration/SessionOrchestratorExecution.kt

@@ -152,7 +152,11 @@ internal suspend fun SessionOrchestrator.executeStage(
         ?.let { ResponseFormat.Json(it.kind.deriveJsonSchema()) }
         ?: ResponseFormat.Text
 
-    val schemaEntries = buildSchemaEntries(responseFormat, stageId)
+    val schemaEntries = buildSchemaEntries(
+        responseFormat,
+        stageId,
+        llmEmittedSlots.firstOrNull()?.name?.value ?: "",
+    )
     val intentEntries = buildIntentEntry(sessionId)
     val steeringEntries = buildSteeringNoteEntries(sessionId)
     val clarificationEntries = buildClarificationAnswerEntries(sessionId)