fix(slack): render mrkdwn + keep scheduling on the local daemon

src/config/profile.ts

@@ -122,7 +122,7 @@ You are the user's digital clone with access to their message history, accumulat
 
 You run inside the Nomos daemon — a self-hosted, locally-running agent. You are NOT Claude.ai's web product.
 
-**Scheduling — IMPORTANT.** You may notice a built-in skill called \`schedule\` (sometimes invoked as \`/schedule\`) in your skill list. **DO NOT USE IT.** That skill creates Anthropic-hosted Routines with a 1-hour minimum interval — it is a different product unrelated to Nomos. **Always** use the \`schedule_task\` MCP tool instead. It accepts any interval ('1m', '5m', '15m', '30m', '1h', etc.) and runs the task inside the local Nomos daemon. There is no minimum interval. Examples:
+**Scheduling — IMPORTANT.** You may notice built-in skills called \`schedule\` (\`/schedule\`) and \`loop\` (\`/loop\`) in your skill list, and built-in tools named \`CronCreate\`, \`RemoteTrigger\`, or \`ScheduleWakeup\`. **DO NOT USE ANY OF THEM.** They create Anthropic-hosted claude.ai Routines (1-hour minimum, results land on the claude.ai dashboard, and they never run in the daemon or show up in the user's settings) — a different product unrelated to Nomos. **Always** use the Nomos MCP tools instead: \`schedule_task\` for one-off or recurring background tasks, and \`loop_create\` for autonomous loops. They accept any interval ('1m', '5m', '15m', '30m', '1h', etc.), have no minimum, and run inside the local Nomos daemon. Examples:
 - "check email every 15 min" → \`schedule_task(schedule='15m', schedule_type='every', prompt='Check unread emails and draft replies')\`
 - "remind me daily at 9am" → \`schedule_task(schedule='0 9 * * *', schedule_type='cron', prompt='...')\`
 
@@ -315,6 +315,13 @@ You can create background tasks that run automatically in the daemon:
 When the user asks for recurring actions (e.g. "check my emails every 15 minutes", "remind me daily"), create a scheduled task instead of suggesting manual checks. Tasks run in the background even between conversations.`,
   );
 
+  // Asking & planning — these tools render proper cards in the app.
+  sections.push(
+    `## Asking & planning
+- **Need the user to decide or supply a missing detail before you can act?** ALWAYS ask through the \`ask_user\` tool — NEVER by writing questions in prose. It shows them tappable choices. Ask the SINGLE most important missing detail first, with 2-4 short options; once they answer, ask the next. Even when several things are unknown (e.g. "book me a dinner reservation" → day, time, party size, place), do NOT dump a numbered list of questions in the chat — pick the one detail that unblocks you and ask it with \`ask_user\`. Only skip the tool when you can reasonably proceed on a sensible default or infer the answer from memory.
+- **Laying out a multi-step plan?** Use the \`TodoWrite\` tool to write the steps as todos — it renders a single tracked plan the user can follow, and you update statuses as you work. Do NOT spin up real scheduled tasks for ephemeral planning steps; \`schedule_task\` is only for things that should actually run on a schedule.`,
+  );
+
   // Permissions
   const permissionsSection = [
     `## Permissions

src/daemon/agent-runtime.ts

@@ -46,11 +46,31 @@ import { resolveMemoryUserId } from "../auth/tenant-context.ts";
  * here so both single-agent and team-runtime call sites stay consistent.
  */
 function getDisallowedTools(): string[] {
-  // The SDK's generic `Workflow` orchestration tool spawns its OWN sub-agents
-  // outside Nomos's team runtime (no memory/persona, an async "notify when done"
-  // model that doesn't fit a single chat turn) and leaks a raw script into the UI.
-  // Block it so "spin up a team" routes to the Nomos-native `delegate_to_team`.
-  const blocked: string[] = ["Workflow"];
+  // Block the SDK's generic orchestration/task built-ins so the agent routes to the
+  // Nomos-native equivalents (which render proper cards + own durable state):
+  //  - `Workflow` spawns sub-agents outside the team runtime + leaks a raw script →
+  //    use `delegate_to_team`.
+  //  - `TaskCreate`/`TaskList`/`TaskUpdate`/`TaskDelete` are the SDK task tracker;
+  //    they render as raw CoT noise instead of a Plan card and create stray tasks →
+  //    use `TodoWrite` for a tracked plan, `schedule_task`/`loop_create` for real ones.
+  //  - `CronCreate`/`CronDelete`/`CronList`/`RemoteTrigger`/`ScheduleWakeup` are what the
+  //    built-in `schedule` + `loop` skills call to create Anthropic-hosted claude.ai
+  //    Routines (1-hour minimum, results land on the claude.ai dashboard, never run in
+  //    the daemon and never show in the user's settings). A prompt warning alone didn't
+  //    stop the agent from reaching for them, so block them outright → the agent must use
+  //    the `schedule_task` / `loop_create` MCP tools, which run locally in the daemon.
+  const blocked: string[] = [
+    "Workflow",
+    "TaskCreate",
+    "TaskList",
+    "TaskUpdate",
+    "TaskDelete",
+    "CronCreate",
+    "CronDelete",
+    "CronList",
+    "RemoteTrigger",
+    "ScheduleWakeup",
+  ];
   if (!FEATURES.bashTool()) {
     blocked.push("Bash", "BashOutput", "KillBash");
   }
@@ -88,16 +108,25 @@ function summarizeToolInput(name: string, input: unknown): string {
     case "Grep":
     case "Glob":
       return str(o.pattern);
+    case "Skill":
+      return str(o.skill) || str(o.name) || str(o.command);
     default: {
-      const pick =
-        str(o.query) || str(o.prompt) || str(o.path) || str(o.message) || str(o.description);
-      if (pick) return pick;
-      try {
-        const j = JSON.stringify(input);
-        return j && j.length > 2 ? (j.length > 120 ? `${j.slice(0, 117)}…` : j) : "";
-      } catch {
-        return "";
-      }
+      // A friendly summary from a known field. NEVER dump raw JSON into the card —
+      // an unrecognized tool with no readable field shows no subtitle instead.
+      return (
+        str(o.query) ||
+        str(o.prompt) ||
+        str(o.path) ||
+        str(o.message) ||
+        str(o.description) ||
+        str(o.skill) ||
+        str(o.name) ||
+        str(o.command) ||
+        str(o.url) ||
+        str(o.content) ||
+        str(o.title) ||
+        ""
+      );
     }
   }
 }

src/daemon/channels/slack.ts

@@ -8,6 +8,7 @@ import SlackBolt from "@slack/bolt";
 import type { ChannelAdapter, IncomingMessage, OutgoingMessage } from "../types.ts";
 import type { DraftManager } from "../draft-manager.ts";
 import { chunkResponse } from "../response-chunker.ts";
+import { markdownToSlackMrkdwn } from "./slack-mrkdwn.ts";
 import { randomUUID } from "node:crypto";
 import { createLogger } from "../../lib/logger.ts";
 
@@ -168,7 +169,9 @@ export class SlackAdapter implements ChannelAdapter {
   async send(message: OutgoingMessage): Promise<void> {
     if (!this.app) return;
     const token = process.env.SLACK_BOT_TOKEN!;
-    const result = chunkResponse(message.content, "slack");
+    // Translate Markdown -> Slack mrkdwn before chunking so the language tag on
+    // fenced code blocks is stripped and **bold**/headings/links render natively.
+    const result = chunkResponse(markdownToSlackMrkdwn(message.content), "slack");
 
     for (const text of result.chunks) {
       await this.app.client.chat.postMessage({
@@ -179,13 +182,14 @@ export class SlackAdapter implements ChannelAdapter {
       });
     }
 
-    // Upload full response as file for very long messages
+    // Upload full response as file for very long messages. The file keeps the
+    // original Markdown -- it renders correctly in editors/viewers as-is.
     if (result.strategy === "file" && result.fullText && result.filename) {
       const uploadArgs = {
         token,
         channel_id: message.channelId,
         filename: result.filename,
-        content: result.fullText,
+        content: message.content,
         title: "Full Response",
         ...(message.threadId ? { thread_ts: message.threadId } : {}),
       };
@@ -203,7 +207,7 @@ export class SlackAdapter implements ChannelAdapter {
     const result = await this.app.client.chat.postMessage({
       token: process.env.SLACK_BOT_TOKEN!,
       channel: channelId,
-      text,
+      text: markdownToSlackMrkdwn(text),
       thread_ts: threadId,
     });
     return result.ts;
@@ -215,7 +219,7 @@ export class SlackAdapter implements ChannelAdapter {
       token: process.env.SLACK_BOT_TOKEN!,
       channel: channelId,
       ts: messageId,
-      text,
+      text: markdownToSlackMrkdwn(text),
     });
   }