feat(discord): add /kill and /stop to interrupt a running agent

.claude-plugin/marketplace.json

@@ -8,7 +8,7 @@
       "name": "claudeclaw",
       "source": "./",
       "description": "Cron-like daemon that runs Claude prompts on a schedule",
-      "version": "1.0.44",
+      "version": "1.0.45",
       "keywords": [
         "cron",
         "heartbeat",

.claude-plugin/plugin.json

@@ -1,5 +1,5 @@
 {
   "name": "claudeclaw",
-  "version": "1.0.44",
+  "version": "1.0.45",
   "description": "Cron-like daemon that runs Claude prompts on a schedule"
 }

src/commands/discord.ts

@@ -1,4 +1,4 @@
-import { ensureProjectClaudeMd, run, runUserMessage, compactCurrentSession, compactCurrentThreadSession, agentDirKey } from "../runner";
+import { ensureProjectClaudeMd, run, runUserMessage, killActive, compactCurrentSession, compactCurrentThreadSession, agentDirKey } from "../runner";
 import { wrapUntrusted } from "../prompt-safety";
 import { isAllowed, isDiscordAuthorized } from "../allowlist";
 import { extractErrorDetail } from "../messaging";
@@ -498,7 +498,7 @@ Rules:
     const { execSync } = await import("node:child_process");
     const input = `${systemPrompt}\n\n---\nUser message: ${text}`;
     const result = execSync(
-      `claude --model claude-sonnet-4-20250514 --print --output-format text`,
+      `claude --model claude-sonnet-5 --print --output-format text`,
       {
         input,
         encoding: "utf-8",
@@ -977,6 +977,18 @@ async function handleMessageCreate(token: string, message: DiscordMessage, skipC
     // Skill routing: detect slash commands and resolve to SKILL.md prompts
     const command = cleanContent.startsWith("/") ? cleanContent.trim().split(/\s+/, 1)[0].toLowerCase() : null;
 
+    if (command === "/kill" || command === "/stop") {
+      // killActive() is global: it kills every in-flight main run across all
+      // Discord channels/threads, not just the caller's. Combined with the
+      // channel-scoped allowlist (#248), a user authorized in one channel can
+      // kill runs in channels they aren't authorized for. Matches Telegram's
+      // existing /kill contract (not a regression); thread-scoped kill is
+      // tracked separately (see #237).
+      const killed = killActive();
+      await sendMessage(config.token, channelId, killed ? "Killed active agent." : "No active agent running.");
+      return;
+    }
+
     let skillContext: string | null = null;
     if (command) {
       try {