diff --git a/packages/junior/src/chat/prompt.ts b/packages/junior/src/chat/prompt.ts
index b373e4d04..562bd4ee3 100644
--- a/packages/junior/src/chat/prompt.ts
+++ b/packages/junior/src/chat/prompt.ts
@@ -1,6 +1,6 @@
 import fs from "node:fs";
 import path from "node:path";
-import { botConfig, getRuntimeMetadata } from "@/chat/config";
+import { botConfig } from "@/chat/config";
 import { TURN_CONTEXT_TAG } from "@/chat/turn-context-tag";
 import {
   listReferenceFiles,
@@ -171,7 +171,7 @@ function formatSkillEntry(skill: SkillMetadata): string[] {
 function formatAvailableSkillsForPrompt(
   skills: SkillMetadata[],
   invocation: SkillInvocation | null,
-): string {
+): string | null {
   const autoSelectable = skills.filter(
     (s) => s.disableModelInvocation !== true,
   );
@@ -184,20 +184,18 @@ function formatAvailableSkillsForPrompt(
 
   const sections: string[] = [];
 
-  // Available skills: model may load these when they match the request.
-  const available = [
-    "<available-skills>",
-    ...(autoSelectable.length > 0
-      ? [
-          "Scan before answering. Load the most specific matching skill; do not answer from memory when a skill fits. If none fits, do not load a skill.",
-        ]
-      : []),
-  ];
-  for (const skill of autoSelectable) {
-    available.push(...formatSkillEntry(skill));
+  if (autoSelectable.length > 0) {
+    // Available skills: model may load these when they match the request.
+    const available = [
+      "<available-skills>",
+      "Scan before answering. Load the most specific matching skill; do not answer from memory when a skill fits. If none fits, do not load a skill.",
+    ];
+    for (const skill of autoSelectable) {
+      available.push(...formatSkillEntry(skill));
+    }
+    available.push("</available-skills>");
+    sections.push(available.join("\n"));
   }
-  available.push("</available-skills>");
-  sections.push(available.join("\n"));
 
   // User-callable skills: model must not auto-select these.
   if (invokedExplicitOnly.length > 0) {
@@ -212,12 +210,12 @@ function formatAvailableSkillsForPrompt(
     sections.push(userCallable.join("\n"));
   }
 
-  return sections.join("\n");
+  return sections.length > 0 ? sections.join("\n") : null;
 }
 
-function formatLoadedSkillsForPrompt(skills: Skill[]): string {
+function formatLoadedSkillsForPrompt(skills: Skill[]): string | null {
   if (skills.length === 0) {
-    return "<loaded-skills>\n</loaded-skills>";
+    return null;
   }
 
   const lines = ["<loaded-skills>"];
@@ -379,23 +377,6 @@ function formatConfigurationLines(
   );
 }
 
-function formatSlackCapabilityNames(
-  capabilities:
-    | {
-        canAddReactions?: boolean;
-        canCreateCanvas?: boolean;
-        canPostToChannel?: boolean;
-      }
-    | undefined,
-): string {
-  const names = [
-    capabilities?.canCreateCanvas ? "canvas_create" : "",
-    capabilities?.canPostToChannel ? "channel_post" : "",
-    capabilities?.canAddReactions ? "reaction_add" : "",
-  ].filter(Boolean);
-  return names.length > 0 ? names.join(", ") : "none";
-}
-
 const HEADER =
   "You are a Slack-based helper assistant. Follow the personality block for voice and tone in every reply. The behavior and output blocks define platform mechanics and override personality only when those mechanics conflict.";
 
@@ -500,32 +481,20 @@ function buildIdentitySection(): string {
 }
 
 function buildRuntimeSection(params: {
-  channelId?: string;
-  fastModelId?: string;
-  modelId?: string;
-  slackCapabilities?: {
-    canAddReactions?: boolean;
-    canCreateCanvas?: boolean;
-    canPostToChannel?: boolean;
-  };
-  thinkingLevel?: string;
-}): string {
+  conversationId?: string;
+  traceId?: string;
+}): string | null {
   const lines = [
-    `- version: ${escapeXml(getRuntimeMetadata().version ?? "unknown")}`,
-    params.modelId ? `- model: ${escapeXml(params.modelId)}` : "",
-    params.fastModelId ? `- fast_model: ${escapeXml(params.fastModelId)}` : "",
-    params.thinkingLevel
-      ? `- thinking: ${escapeXml(params.thinkingLevel)}`
+    params.conversationId
+      ? `- gen_ai.conversation.id: ${escapeXml(params.conversationId)}`
       : "",
-    params.channelId ? "- channel: slack" : "",
-    params.channelId
-      ? `- slack_capabilities: ${escapeXml(
-          formatSlackCapabilityNames(params.slackCapabilities),
-        )}`
-      : "",
-    `- sandbox_workspace: ${escapeXml(SANDBOX_WORKSPACE_ROOT)}`,
+    params.traceId ? `- trace_id: ${escapeXml(params.traceId)}` : "",
   ].filter(Boolean);
 
+  if (lines.length === 0) {
+    return null;
+  }
+
   return renderTagBlock("runtime", lines.join("\n"));
 }
 
@@ -535,7 +504,7 @@ function buildContextSection(params: {
   configuration?: Record<string, unknown>;
   invocation: SkillInvocation | null;
   turnState?: "fresh" | "resumed";
-}): string {
+}): string | null {
   const blocks: string[][] = [];
 
   if (JUNIOR_WORLD) {
@@ -593,6 +562,10 @@ function buildContextSection(params: {
   }
 
   const body = blocks.map((block) => block.join("\n")).join("\n\n");
+  if (!body) {
+    return null;
+  }
+
   return renderTagBlock("context", body);
 }
 
@@ -602,12 +575,20 @@ function buildCapabilitiesSection(params: {
   activeMcpCatalogs: ActiveMcpCatalogSummary[];
   invocation: SkillInvocation | null;
   toolGuidance?: ToolPromptContext[];
-}): string {
+}): string | null {
   const blocks: string[] = [];
-  blocks.push(
-    formatAvailableSkillsForPrompt(params.availableSkills, params.invocation),
+  const availableSkills = formatAvailableSkillsForPrompt(
+    params.availableSkills,
+    params.invocation,
   );
-  blocks.push(formatLoadedSkillsForPrompt(params.activeSkills));
+  if (availableSkills) {
+    blocks.push(availableSkills);
+  }
+
+  const loadedSkills = formatLoadedSkillsForPrompt(params.activeSkills);
+  if (loadedSkills) {
+    blocks.push(loadedSkills);
+  }
 
   const activeCatalogs = formatActiveMcpCatalogsForPrompt(
     params.activeMcpCatalogs,
@@ -626,6 +607,10 @@ function buildCapabilitiesSection(params: {
     blocks.push(renderTagBlock("providers", providerCatalog));
   }
 
+  if (blocks.length === 0) {
+    return null;
+  }
+
   return renderTagBlock("capabilities", blocks.join("\n\n"));
 }
 
@@ -635,15 +620,8 @@ type TurnContextPromptInput = {
   activeMcpCatalogs?: ActiveMcpCatalogSummary[];
   toolGuidance?: ToolPromptContext[];
   runtime?: {
-    channelId?: string;
-    fastModelId?: string;
-    modelId?: string;
-    slackCapabilities?: {
-      canAddReactions?: boolean;
-      canCreateCanvas?: boolean;
-      canPostToChannel?: boolean;
-    };
-    thinkingLevel?: string;
+    conversationId?: string;
+    traceId?: string;
   };
   invocation: SkillInvocation | null;
   requester?: {
@@ -702,7 +680,7 @@ export function buildTurnContextPrompt(params: TurnContextPromptInput): string {
     }),
     buildRuntimeSection(params.runtime ?? {}),
     `</${TURN_CONTEXT_TAG}>`,
-  ];
+  ].filter((section): section is string => Boolean(section));
 
   return sections.join("\n\n");
 }
diff --git a/packages/junior/src/chat/respond-helpers.ts b/packages/junior/src/chat/respond-helpers.ts
index baeb04326..2afb363da 100644
--- a/packages/junior/src/chat/respond-helpers.ts
+++ b/packages/junior/src/chat/respond-helpers.ts
@@ -149,62 +149,28 @@ export function summarizeMessageText(text: string): string {
 }
 
 /**
- * Wrap the current user turn with self-describing marker blocks: background
- * first, current instruction last. Ordering matches long-context attention
- * guidance for Sonnet and GPT-5.
+ * Put prior thread text before the current instruction when no Pi history
+ * exists. These are top-level sibling blocks in the user message.
  */
 export function buildUserTurnText(
   userInput: string,
   conversationContext?: string,
-  metadata?: {
-    sessionContext?: { conversationId?: string };
-    turnContext?: { traceId?: string };
-  },
 ): string {
   const trimmedContext = conversationContext?.trim();
-  const conversationId = metadata?.sessionContext?.conversationId;
-  const traceId = metadata?.turnContext?.traceId;
 
-  if (!trimmedContext && !conversationId && !traceId) {
+  if (!trimmedContext) {
     return userInput;
   }
 
-  const sections: string[] = [];
-
-  if (trimmedContext) {
-    sections.push(
-      "<thread-background>",
-      trimmedContext,
-      "</thread-background>",
-      "",
-    );
-  }
-
-  if (conversationId) {
-    sections.push(
-      "<session-context>",
-      `- gen_ai.conversation.id: ${conversationId}`,
-      "</session-context>",
-      "",
-    );
-  }
-
-  if (traceId) {
-    sections.push(
-      "<turn-context>",
-      `- trace_id: ${traceId}`,
-      "</turn-context>",
-      "",
-    );
-  }
-
-  sections.push(
-    '<current-instruction priority="highest">',
+  return [
+    "<thread-background>",
+    trimmedContext,
+    "</thread-background>",
+    "",
+    "<current-instruction>",
     userInput,
     "</current-instruction>",
-  );
-
-  return sections.join("\n");
+  ].join("\n");
 }
 
 /** Encode a non-image attachment as base64 XML for the prompt. */
diff --git a/packages/junior/src/chat/respond.ts b/packages/junior/src/chat/respond.ts
index ee76f2796..d94152749 100644
--- a/packages/junior/src/chat/respond.ts
+++ b/packages/junior/src/chat/respond.ts
@@ -556,10 +556,6 @@ export async function generateAssistantReply(
     const userTurnText = buildUserTurnText(
       userInput,
       promptConversationContext,
-      {
-        sessionContext: { conversationId: sessionConversationId },
-        turnContext: { traceId: getActiveTraceId() },
-      },
     );
     const { routerBlocks, userContentParts } = buildUserTurnInput({
       omittedImageAttachmentCount: context.omittedImageAttachmentCount ?? 0,
@@ -770,11 +766,8 @@ export async function generateAssistantReply(
       activeMcpCatalogs,
       toolGuidance,
       runtime: {
-        channelId: toolChannelId,
-        fastModelId: botConfig.fastModelId,
-        modelId: botConfig.modelId,
-        slackCapabilities: channelCapabilities,
-        thinkingLevel: thinkingSelection.thinkingLevel,
+        conversationId: spanContext.conversationId,
+        traceId: getActiveTraceId(),
       },
       invocation: skillInvocation,
       requester: context.requester,
diff --git a/packages/junior/tests/unit/misc/respond-helpers-user-turn.test.ts b/packages/junior/tests/unit/misc/respond-helpers-user-turn.test.ts
index b0a2ddc43..47d4b12e7 100644
--- a/packages/junior/tests/unit/misc/respond-helpers-user-turn.test.ts
+++ b/packages/junior/tests/unit/misc/respond-helpers-user-turn.test.ts
@@ -5,4 +5,18 @@ describe("buildUserTurnText", () => {
   it("returns raw input when no context or metadata is provided", () => {
     expect(buildUserTurnText("hello")).toBe("hello");
   });
+
+  it("keeps only causal thread context around the current instruction", () => {
+    expect(buildUserTurnText("what now?", "alice: budget is due Friday")).toBe(
+      [
+        "<thread-background>",
+        "alice: budget is due Friday",
+        "</thread-background>",
+        "",
+        "<current-instruction>",
+        "what now?",
+        "</current-instruction>",
+      ].join("\n"),
+    );
+  });
 });
diff --git a/packages/junior/tests/unit/prompt.test.ts b/packages/junior/tests/unit/prompt.test.ts
index 3741b3719..d32e36558 100644
--- a/packages/junior/tests/unit/prompt.test.ts
+++ b/packages/junior/tests/unit/prompt.test.ts
@@ -18,9 +18,8 @@ describe("prompt builders", () => {
       invocation: null,
       requester: { userId: "U_ALPHA" },
       runtime: {
-        channelId: "C_ALPHA",
-        modelId: "model-alpha",
-        thinkingLevel: "medium",
+        conversationId: "conversation-alpha",
+        traceId: "trace-alpha",
       },
       turnState: "fresh",
     });
@@ -40,9 +39,7 @@ describe("prompt builders", () => {
       invocation: null,
       requester: { userId: "U_BETA" },
       runtime: {
-        channelId: "C_BETA",
-        modelId: "model-beta",
-        thinkingLevel: "high",
+        conversationId: "conversation-beta",
       },
       turnState: "resumed",
     });
@@ -55,6 +52,17 @@ describe("prompt builders", () => {
     expect(firstTurnContext).not.toContain("<assistant>");
     expect(firstTurnContext).not.toContain("<thread-participants>");
     expect(firstTurnContext).toContain("<requester>");
+    expect(firstTurnContext).toContain(
+      "- gen_ai.conversation.id: conversation-alpha",
+    );
+    expect(firstTurnContext).toContain("- trace_id: trace-alpha");
+    expect(firstTurnContext).not.toContain("- model:");
+    expect(firstTurnContext).not.toContain("- fast_model:");
+    expect(firstTurnContext).not.toContain("- channel:");
+    expect(firstTurnContext).not.toContain("- slack_capabilities:");
+    expect(firstTurnContext).not.toContain("- thinking:");
+    expect(firstTurnContext).not.toContain("- sandbox_workspace:");
+    expect(firstSystemPrompt).not.toContain("trace-alpha");
     expect(buildSystemPrompt()).toBe(firstSystemPrompt);
   });
 
@@ -68,6 +76,9 @@ describe("prompt builders", () => {
     });
 
     expect(turnContext).not.toContain("<requester>");
+    expect(turnContext).not.toContain("<context>");
+    expect(turnContext).not.toContain("<capabilities>");
+    expect(turnContext).not.toContain("<runtime>");
   });
 
   it("puts tool guidance in turn context, not the static system prompt", () => {
diff --git a/packages/junior/tests/unit/runtime/respond-mcp-progressive-loading.test.ts b/packages/junior/tests/unit/runtime/respond-mcp-progressive-loading.test.ts
index 1ab3bde7c..40d43b0e9 100644
--- a/packages/junior/tests/unit/runtime/respond-mcp-progressive-loading.test.ts
+++ b/packages/junior/tests/unit/runtime/respond-mcp-progressive-loading.test.ts
@@ -17,6 +17,7 @@ const {
   omitFinalAssistantAfterTool,
   pushPreToolAssistantMessage,
   promptCallCount,
+  promptMessages,
   promptSeedMessages,
   recordToolResultMessage,
   resumeTurnContextCounts,
@@ -42,6 +43,7 @@ const {
   loadSkillsByNameMock: vi.fn(),
   omitFinalAssistantAfterTool: { value: false },
   promptCallCount: { value: 0 },
+  promptMessages: [] as unknown[],
   promptSeedMessages: [] as unknown[][],
   pushPreToolAssistantMessage: { value: false },
   recordToolResultMessage: { value: false },
@@ -143,6 +145,7 @@ vi.mock("@earendil-works/pi-agent-core", () => {
     async prompt(message: unknown) {
       promptCallCount.value += 1;
       this.aborted = false;
+      promptMessages.push(message);
       promptSeedMessages.push([...this.state.messages]);
       this.state.messages.push(message);
 
@@ -564,6 +567,7 @@ describe("generateAssistantReply progressive MCP loading", () => {
     loadSkillsByNameMock.mockReset();
     omitFinalAssistantAfterTool.value = false;
     promptCallCount.value = 0;
+    promptMessages.length = 0;
     promptSeedMessages.length = 0;
     pushPreToolAssistantMessage.value = false;
     recordToolResultMessage.value = false;
@@ -759,6 +763,12 @@ describe("generateAssistantReply progressive MCP loading", () => {
     });
 
     expect(promptSeedMessages[0]).toEqual(priorMessages);
+    expect(JSON.stringify(promptMessages[0])).not.toContain(
+      "duplicated prior transcript",
+    );
+    expect(JSON.stringify(promptMessages[0])).not.toContain(
+      "<thread-background>",
+    );
   });
 
   it("parks for auth when MCP auth is requested during a tool call", async () => {