fix: workflow inference chain — surface llama-server errors, user-turn-last, bundle-relative prompts

- LlamaCppInferenceProvider: check HTTP status and surface the real error body instead of masking it as a ChatCompletionResponse deserialization failure
- PromptRenderer: render the live conversation layer (L1) last so the user turn is the final message (fixes Qwen 'No user query found' 500 when L3 recall is present)
- TomlWorkflowLoader: resolve stage prompts relative to the workflow file's own dir (bundle), CWD-independent; config-dir fallback for globals
- SessionOrchestrator: hard-fail a stage whose declared prompt can't resolve (was a silent user-less request)
- move healthcheck prompts next to the example workflow (examples/workflows/prompts/)

core/inference/src/main/kotlin/com/correx/core/inference/PromptRenderer.kt

@@ -12,6 +12,14 @@ data class ChatMessage(
 )
 
 object PromptRenderer {
+    // L1 (live conversation, ends with the current user turn) must render last so
+    // the model template sees the user query as the final message. Background/memory
+    // layers (L2, L3, L4) are injected before it.
+    private val nonL0RenderOrder: Comparator<ContextLayer> = Comparator { a, b ->
+        val priority = { layer: ContextLayer -> if (layer == ContextLayer.L1) Int.MAX_VALUE else layer.ordinal }
+        priority(a) - priority(b)
+    }
+
     fun render(contextPack: ContextPack): List<ChatMessage> {
         val sorted = contextPack.layers.entries.sortedBy { it.key.ordinal }
         val systemContent = sorted
@@ -19,19 +27,22 @@ object PromptRenderer {
             .flatMap { it.value }
             .joinToString("\n\n") { it.content }
             .takeIf { it.isNotBlank() }
-        val conversationMessages = sorted.filter { it.key != ContextLayer.L0 }.flatMap { (_, entries) ->
-            entries.map { entry ->
-                ChatMessage(
-                    role = when (entry.role) {
-                        EntryRole.SYSTEM -> "system"
-                        EntryRole.ASSISTANT -> "assistant"
-                        EntryRole.TOOL -> "tool"
-                        EntryRole.USER -> "user"
-                    },
-                    content = entry.content,
-                )
+        val conversationMessages = contextPack.layers.entries
+            .filter { it.key != ContextLayer.L0 }
+            .sortedWith(Comparator.comparing({ it.key }, nonL0RenderOrder))
+            .flatMap { (_, entries) ->
+                entries.map { entry ->
+                    ChatMessage(
+                        role = when (entry.role) {
+                            EntryRole.SYSTEM -> "system"
+                            EntryRole.ASSISTANT -> "assistant"
+                            EntryRole.TOOL -> "tool"
+                            EntryRole.USER -> "user"
+                        },
+                        content = entry.content,
+                    )
+                }
             }
-        }
         val messages = buildList {
             systemContent?.let { add(ChatMessage("system", it)) }
             addAll(conversationMessages)

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

@@ -222,7 +222,7 @@ abstract class SessionOrchestrator(
             } ?: emptyList()
         val promptEntries = stageConfig.metadata["prompt"]
             ?.let { path ->
-                runCatching { promptResolver.resolve(path) }
+                val resolvedText = runCatching { promptResolver.resolve(path) }
                     .onFailure {
                         log.error(
                             "[SessionOrchestrator] stage=${stageId.value}: " +
@@ -230,9 +230,11 @@ abstract class SessionOrchestrator(
                         )
                     }
                     .getOrNull()
-            }
-            ?.takeIf { it.isNotBlank() }
-            ?.let { text ->
+                val text = resolvedText?.takeIf { it.isNotBlank() }
+                    ?: error(
+                        "[SessionOrchestrator] stage=${stageId.value}: " +
+                                "declared prompt '$path' could not be resolved",
+                    )
                 listOf(
                     ContextEntry(
                         id = ContextEntryId(UUID.randomUUID().toString()),

infrastructure/inference/llama_cpp/src/main/kotlin/com/correx/infrastructure/inference/llama/cpp/LlamaCppInferenceProvider.kt

@@ -21,6 +21,7 @@ import io.ktor.client.request.accept
 import io.ktor.client.request.get
 import io.ktor.client.request.post
 import io.ktor.client.request.setBody
+import io.ktor.client.statement.bodyAsText
 import io.ktor.http.ContentType
 import io.ktor.http.contentType
 import io.ktor.serialization.kotlinx.json.json
@@ -106,11 +107,16 @@ class LlamaCppInferenceProvider(
         val encoded = json.encodeToString(body)
         log.debug("sending request to llm: {}", encoded)
 
-        val response = httpClient.post("$baseUrl/v1/chat/completions") {
+        val httpResponse = httpClient.post("$baseUrl/v1/chat/completions") {
             contentType(ContentType.Application.Json)
             accept(ContentType.Application.Json)
             setBody(encoded)
-        }.body<ChatCompletionResponse>()
+        }
+        if (httpResponse.status.value !in 200..299) {
+            val errorBody = httpResponse.bodyAsText()
+            error("llama-server returned ${httpResponse.status.value} ${httpResponse.status.description}: $errorBody")
+        }
+        val response = httpResponse.body<ChatCompletionResponse>()
 
         log.debug("got response from llm: {}", response)
 

infrastructure/workflow/src/main/kotlin/com/correx/infrastructure/workflow/TomlWorkflowLoader.kt

@@ -15,6 +15,7 @@ import com.fasterxml.jackson.dataformat.toml.TomlMapper
 import com.fasterxml.jackson.module.kotlin.kotlinModule
 import com.fasterxml.jackson.module.kotlin.readValue
 import java.nio.file.Path
+import kotlin.io.path.exists
 import kotlin.io.path.readText
 
 private data class WorkflowFile(
@@ -67,10 +68,17 @@ class TomlWorkflowLoader(
         val raw = path.readText()
         val file = runCatching { mapper.readValue<WorkflowFile>(raw) }
             .getOrElse { throw WorkflowValidationException("Failed to parse workflow TOML: ${it.message}") }
-        return file.toWorkflowGraph()
+        return file.toWorkflowGraph(path.parent)
     }
 
-    private fun WorkflowFile.toWorkflowGraph(): WorkflowGraph {
+    private fun resolvePromptPath(raw: String, workflowDir: Path?): String {
+        val p = Path.of(raw)
+        if (p.isAbsolute) return raw
+        val candidate = workflowDir?.resolve(raw)
+        return if (candidate != null && candidate.exists()) candidate.toString() else raw
+    }
+
+    private fun WorkflowFile.toWorkflowGraph(workflowDir: Path?): WorkflowGraph {
         val startId = StageId(start)
         val stageMap = stages.associate { s ->
             StageId(s.id) to StageConfig(
@@ -84,8 +92,8 @@ class TomlWorkflowLoader(
                 tokenBudget = s.tokenBudget,
                 maxRetries = s.maxRetries,
                 metadata = buildMap {
-                    s.prompt?.let { put("prompt", it) }
-                    s.systemPrompt?.let { put("systemPrompt", it) }
+                    s.prompt?.let { put("prompt", resolvePromptPath(it, workflowDir)) }
+                    s.systemPrompt?.let { put("systemPrompt", resolvePromptPath(it, workflowDir)) }
                 },
             )
         }