switchroom 0.16.47 → 0.17.0

package/skills/switchroom-architecture/telegram.md

@@ -2,12 +2,11 @@
 
 Switchroom ships an enhanced `switchroom-telegram` MCP plugin that replaces the official marketplace plugin. It is the default — no configuration needed.
 
-## 10 MCP tools
+## 9 MCP tools
 
 | Tool | What it does |
 |------|-------------|
-| `reply` | Send text, photos, or documents. Supports threading, topic routing, file attachments. |
-| `stream_reply` | Edit a single message in place as work progresses (~1/sec throttle). Use for long tasks to avoid chat spam. |
+| `reply` | Send text, photos, or documents — the single final-answer tool. Chunks anything over Telegram's 4096-char limit. Supports threading, topic routing, file attachments. |
 | `react` | Add emoji reactions to messages (Telegram whitelist: 👍 👎 ❤️ 🔥 👀 🎉 etc). |
 | `edit_message` | Update a previously sent message. Edits are silent (no push notification). |
 | `delete_message` | Remove a bot-sent message (48h Telegram API limit). |
@@ -37,19 +36,13 @@ A local SQLite database (`telegram/history.db`) records every message. After a C
 
 History survives process restarts and session resets.
 
-## stream_reply pattern
+## Progress while working
 
-For tasks taking more than ~5 seconds:
-
-```
-1. stream_reply(chat_id, "Reading the file...", done=false)  ← creates message
-2. stream_reply(chat_id, "Reading the file...\nParsing...", done=false)  ← edits in place
-3. stream_reply(chat_id, "Done! Here's the result: ...", done=true)  ← locks
-```
-
-Pass the **full current text** on each call (not a delta). The plugin throttles to ~1/sec.
-
-After `done=true`, send a separate `reply` if you want a push notification to the user's device (edits are silent).
+For long tasks you do not need to narrate progress by editing a message. The
+plugin renders an event-driven progress card (Plan → Run → Done with live tool
+bullets, elapsed time, and status emoji) for free while the turn is in-flight.
+Send the final answer once, with `reply` — it chunks anything over Telegram's
+4096-char limit. For an explicit mid-turn check-in use `progress_update`.
 
 ## Formatting
 

package/skills/switchroom-cli/SKILL.md

@@ -10,7 +10,7 @@ This skill is the reference for running `switchroom` CLI commands against existi
 
 **Four commands to know:**
 - `switchroom update` — full operator path: pulls images + applies config + recreates containers + runs doctor (since v0.7.8 / #918). What you want 95% of the time.
-- `switchroom apply` — config-only reconcile: refresh per-agent scaffolds and (re)write `~/.switchroom/compose/docker-compose.yml` without touching running containers. Use when you want to inspect the generated compose before bringing the fleet up yourself.
+- `switchroom apply` — **host/operator-only**, for structural changes: refresh per-agent scaffolds and (re)write `~/.switchroom/compose/docker-compose.yml`. Its full per-agent scaffold **cannot run from inside an agent container** (no vault at the container HOME, no `docker compose` v2 plugin) — that's correct by construction, not a bug. Version rolls do NOT need it: drive the **hostd rollout** (`mcp__hostd__rollout`), which runs a `--compose-only` apply plus a per-agent restart-reconcile that refreshes each agent's templates automatically. Reserve a host-side `sudo switchroom apply` for compose regeneration / new-agent scaffolding.
 - `switchroom restart [agent]` — bounces a stuck or wedged agent
 - `switchroom version` — shows what's running (versions + health summary)
 
@@ -59,7 +59,7 @@ switchroom update --rebuild      # source-checkout users: also git pull + npm bu
 
 `switchroom update` is the operator path. The CLI self-elevates via sudo internally for the per-agent scaffold dirs that need root — no need for `sudo HOME=… PATH=…` incantations.
 
-If you only need the config-reconcile half without restarting agents, `switchroom apply` writes `~/.switchroom/compose/docker-compose.yml` and refreshes per-agent scaffolds without touching running containers. The operator runs the docker bring-up themselves.
+If you only need the config-reconcile half without restarting agents, `switchroom apply` (run **on the host by the operator**) writes `~/.switchroom/compose/docker-compose.yml` and refreshes per-agent scaffolds without touching running containers. The operator runs the docker bring-up themselves. Note this full apply cannot run from inside an agent container by construction — to roll the fleet to a new version from an agent, use the hostd rollout (`mcp__hostd__rollout`) instead of a full apply.
 
 From inside an agent's Telegram DM, the same flow is available as `/upgradestatus` (read-only) and `/update apply` (admin-gated).
 
@@ -265,12 +265,11 @@ If the user asks whether scheduled runs were missed during downtime: the schedul
 
 ## Telegram plugin reference — "what MCP tools", "how does reply work"
 
-The `switchroom-telegram` plugin is an enhanced fork of the official Telegram MCP plugin and is the default for all switchroom agents. It exposes **10 MCP tools** (all prefixed `mcp__switchroom-telegram__`):
+The `switchroom-telegram` plugin is an enhanced fork of the official Telegram MCP plugin and is the default for all switchroom agents. It exposes **9 MCP tools** (all prefixed `mcp__switchroom-telegram__`):
 
 | Tool | Purpose |
 |---|---|
-| `reply` | Send a text/photo message, with optional `reply_to` for threaded quotes |
-| `stream_reply` | Incrementally stream a long reply (edits the same message as tokens arrive) |
+| `reply` | Send a text/photo message — the single final-answer tool; chunks anything over Telegram's 4096-char limit. Optional `reply_to` for threaded quotes |
 | `react` | Emoji reaction on an inbound or outbound message |
 | `edit_message` | Modify an earlier bot message's text |
 | `delete_message` | Remove an earlier bot message |

package/skills/telegram-test-harness/SKILL.md

@@ -191,4 +191,4 @@ Before writing the test, ask:
 - `telegram-plugin/tests/fake-bot-api.test.ts` — meta-test of the fake;
   read first when adding new fake-bot capabilities
 - `telegram-plugin/tests/streaming-e2e.test.ts` — worked example of a
-  larger end-to-end test (PTY → stream_reply → done)
+  larger end-to-end test (PTY → internal stream-reply handler → done)

package/telegram-plugin/README.md

@@ -9,10 +9,12 @@ and substantially extended for long-running, multi-agent deployments.
 The upstream plugin handles the basic send/receive wire protocol. This fork
 adds the ergonomics and reliability that an always-on agent fleet needs:
 
-- **Streaming replies** — `stream_reply` edits a single message in place as
-  work progresses (~1/sec throttle), so users see live progress instead of
-  silent gaps followed by a wall-of-text. Optional `lane` parameter lets each
-  lane (e.g. `thinking` vs default `answer`) get its own message per chat+thread.
+- **Live progress card** — an event-driven progress card (Plan → Run → Done
+  with live tool bullets, elapsed time, and status emoji) edits a single
+  message in place as work progresses, so users see live progress instead of
+  silent gaps followed by a wall-of-text. This is rendered by the plugin for
+  free while the turn is in-flight — the model just sends the final answer once
+  via `reply`.
 - **Status reactions** — emoji lifecycle (👀 queued → 🤔 thinking → 👨‍💻 tool
   use → 🔥 streaming → 👍 done) on the user's own message gives "I'm working"
   feedback for free, plus stall watchdogs (🥱 30s idle, 😨 90s).
@@ -36,7 +38,7 @@ adds the ergonomics and reliability that an always-on agent fleet needs:
 - **Switchroom slash-commands** — `/agents`, `/restart`, `/logs`, `/memory`,
   `/grant`, `/dangerous`, `/permissions`, `/reconcile` etc., handled by the
   plugin without consuming Claude Code tokens.
-- **12 MCP tools** — `reply`, `stream_reply`, `react`, `edit_message`,
+- **11 MCP tools** — `reply`, `react`, `edit_message`,
   `delete_message`, `forward_message`, `pin_message`, `send_typing`,
   `download_attachment`, `get_recent_messages`, `send_checklist`,
   `update_checklist` (the latter two ship native Telegram checklists,
@@ -196,40 +198,27 @@ When an inbound message is received, the plugin immediately reacts with an emoji
 
 Set to an empty string `""` to disable. Only Telegram's fixed emoji whitelist is accepted (👍 👎 ❤ 🔥 👀 🎉 etc). A typing indicator is also sent automatically.
 
-### `stream_reply` tool (preferred for multi-step work)
+### Progress while working
 
-Sends or updates a streaming reply that edits one message in-place rather
-than sending many. Call repeatedly during long tasks with full snapshots of
-the current message; the plugin throttles edits to ~1/sec to respect
-Telegram's rate limit. Set `done=true` on the final call to lock the
-message.
-
-| Parameter | Required | Description |
-|-----------|----------|-------------|
-| `chat_id` | yes | Target chat ID |
-| `text` | yes | Full text snapshot (NOT a delta — pass the complete current content each call) |
-| `done` | no | `true` on final call. After `done=true` the stream is locked and further calls are no-ops. Default `false`. |
-| `format` | no | `"html"` (default), `"markdownv2"`, or `"text"` |
-| `lane` | no | Optional lane name. Each lane gets its own Telegram message per chat+thread. Use `lane: "thinking"` to surface reasoning progress alongside the main answer stream. Omit for the default answer lane. |
-| `message_thread_id` | no | Forum topic thread ID (auto-applied from the last inbound message if not specified) |
-
-Hard-stops at 4096 chars (Telegram message limit). On edit-404 (the message
-we're editing was deleted), the plugin sends a fresh message and continues
-the stream against the new id. A short `idleMs` pre-send debounce coalesces
-back-to-back snapshots before the first wire send, avoiding a redundant
-edit when several updates arrive in the same tick.
+For long tasks the model does **not** need to narrate progress or drive a
+streaming message itself. The plugin renders an event-driven progress card
+(Plan → Run → Done with live tool bullets, elapsed time, and status emoji)
+for free while the turn is in-flight. Send the final answer once, with
+`reply` — it chunks anything over Telegram's 4096-char limit. (The retired
+`stream_reply` tool was a redundant, worse alias of `reply`; the internal
+progress-card streaming that drove it is preserved and is what renders the
+live card.)
 
 ### Manual streaming progress via `edit_message`
 
-If you need finer control than `stream_reply` offers, you can drive the
-edit loop yourself:
+If you want to drive an in-place edit loop yourself:
 
 1. Send an initial "thinking..." message with `reply` — note the returned `message_id`
 2. Call `edit_message` with updated text as work progresses (edits are silent — no push notification)
 3. Call `send_typing` between steps to keep the typing indicator alive (it expires after ~5s)
 4. When done, send a **new** `reply` so the user's device pings with a push notification
 
-In most cases `stream_reply` is simpler and is the recommended path.
+In most cases the automatic progress card is simpler and is the recommended path.
 
 ### `send_typing` tool
 

package/telegram-plugin/bridge/bridge.ts

@@ -75,7 +75,7 @@ const mcp = new Server(
     instructions: [
       'The sender reads Telegram, not this session. Anything you want them to see must go through the reply tool — your transcript output never reaches their chat.',
       '',
-      'Messages from Telegram arrive as <channel source="telegram" chat_id="..." message_id="..." user="..." ts="...">. If the tag has an image_path attribute, Read that file — it is a photo the sender attached. If the tag has attachment_file_id, call download_attachment with that file_id to fetch the file, then Read the returned path. A single message may carry SEVERAL attachments (a forwarded album or a text+multi-image burst): when attachment_count is set (>1), also handle the numbered siblings — image_path_2, image_path_3, … (Read each) and attachment_file_id_2, attachment_file_id_3, … (download_attachment each). Process every one, not just the first. Reply with the reply tool — pass chat_id back. The reply and stream_reply tools quote-reply to the latest inbound user message by default, so you do NOT need to pass reply_to for normal responses. Pass reply_to (a message_id) only when quoting a specific earlier message, or pass quote:false to send a bare (non-quoted) message.',
+      'Messages from Telegram arrive as <channel source="telegram" chat_id="..." message_id="..." user="..." ts="...">. If the tag has an image_path attribute, Read that file — it is a photo the sender attached. If the tag has attachment_file_id, call download_attachment with that file_id to fetch the file, then Read the returned path. A single message may carry SEVERAL attachments (a forwarded album or a text+multi-image burst): when attachment_count is set (>1), also handle the numbered siblings — image_path_2, image_path_3, … (Read each) and attachment_file_id_2, attachment_file_id_3, … (download_attachment each). Process every one, not just the first. Reply with the reply tool — pass chat_id back. The reply tool quote-replies to the latest inbound user message by default, so you do NOT need to pass reply_to for normal responses. Pass reply_to (a message_id) only when quoting a specific earlier message, or pass quote:false to send a bare (non-quoted) message.',
       '',
       'reply accepts file paths (files: ["/abs/path.png"]) for attachments. Use react to add emoji reactions, edit_message for interim progress updates, and delete_message when you need to truly remove a message (prefer edit_message if you just want to change text — delete is for retraction). Edits don\'t trigger push notifications — when a long task completes, send a new reply so the user\'s device pings. Use send_typing to show a typing indicator during long operations. Use pin_message to pin important outputs. Use forward_message to quote/resurface earlier messages.',
       '',
@@ -134,46 +134,6 @@ const TOOL_SCHEMAS = [
       required: ['chat_id', 'text'],
     },
   },
-  {
-    name: 'stream_reply',
-    description:
-      'Post the final answer for this turn. The plugin renders an event-driven progress card (Plan → Run → Done with live tool bullets, elapsed time, and status emoji) for free while the turn is in-flight, so you do not need to narrate intermediate progress. Call `stream_reply` exactly once per turn with done=true and the complete answer text. Hard cap is 32768 chars (the rich-message wire limit) — longer text is dropped by a defensive guard, so use `reply` for anything that long (it chunks). Calling with done=false is an error in this environment (the progress card already owns the mid-turn surface). inline_keyboard adds tappable buttons under the final message — see `reply` for shape and constraints.',
-    inputSchema: {
-      type: 'object',
-      properties: {
-        chat_id: { type: 'string' },
-        text: { type: 'string', description: 'Full text snapshot. NOT a delta — pass the complete current content each call.' },
-        done: { type: 'boolean', description: 'Must be true. Posts this text as the final answer for the turn and locks the message.' },
-        message_thread_id: { type: 'string', description: 'Forum topic thread ID. Auto-applied from the last inbound message if not specified.' },
-        origin_turn_id: { type: 'string', description: 'In a forum supergroup, pass back the origin_turn_id attribute from the <channel> message you are answering. It pins the reply to that message\'s topic even if another topic\'s turn started meanwhile. Omit in DMs / single-topic chats.' },
-        format: { type: 'string', enum: ['html', 'markdownv2', 'text'], description: "Rendering mode. 'html' (default) converts markdown to Telegram HTML." },
-        reply_to: { type: 'string', description: 'Message ID to quote-reply to. Overrides the default (latest inbound).' },
-        quote: { type: 'boolean', description: 'Opt out of the default quote-reply behavior. Default: true. Ignored when reply_to is explicitly set.' },
-        protect_content: { type: 'boolean', description: 'When true, Telegram prevents the message from being forwarded or saved.' },
-        quote_text: { type: 'string', description: 'Surgical quote: specific text to highlight from the reply_to message. Requires reply_to.' },
-        disable_notification: { type: 'boolean', description: 'When true, the INITIAL message send is silent (no device ping). Has no effect on subsequent edits — Telegram never pings on editMessageText. Default false. Use for mid-turn stream starts you do not want to ping; omit on the final answer.' },
-        inline_keyboard: {
-          type: 'array',
-          description: '2D array of tappable buttons under the final message. Same shape and constraints as `reply.inline_keyboard` — each button has `text` and EXACTLY ONE of `url` or `callback_data`, plus optional `ack_text` (custom tap-toast; default "✓ received") and `single_use` (default true; set false to keep the keyboard tappable after a tap). Tap on a callback_data button is delivered to this agent as an inbound channel event with meta.button_callback_data set.',
-          items: {
-            type: 'array',
-            items: {
-              type: 'object',
-              properties: {
-                text: { type: 'string' },
-                url: { type: 'string' },
-                callback_data: { type: 'string' },
-                ack_text: { type: 'string', description: 'Toast text shown on tap. Default "✓ received".' },
-                single_use: { type: 'boolean', description: 'Default true. Set false to keep the keyboard tappable after this button is tapped.' },
-              },
-              required: ['text'],
-            },
-          },
-        },
-      },
-      required: ['chat_id', 'text'],
-    },
-  },
   {
     name: 'react',
     description: 'Add an emoji reaction to a Telegram message. Telegram only accepts a fixed whitelist (👍 👎 ❤ 🔥 👀 🎉 etc) — non-whitelisted emoji will be rejected.',

package/telegram-plugin/bridge/tool-filter.ts

@@ -30,13 +30,12 @@ export interface NamedTool {
 }
 
 /**
- * Hot tools pinned loaded — must never defer. The reply path
- * (reply/stream_reply) plus the frequently-used early-turn ops. Everything
- * NOT in this set defers under tool-search.
+ * Hot tools pinned loaded — must never defer. The reply path plus the
+ * frequently-used early-turn ops. Everything NOT in this set defers under
+ * tool-search.
  */
 export const ALWAYS_LOAD_TOOLS: ReadonlySet<string> = new Set([
   'reply',
-  'stream_reply',
   'get_recent_messages',
   'react',
   'edit_message',

package/telegram-plugin/dist/bridge/bridge.js

@@ -23169,7 +23169,7 @@ function findActiveSessionFile(projectsDir) {
 }
 function parseChannelMeta(content) {
   const grab = (key) => {
-    const m = content.match(new RegExp(`${key}="([^"]+)"`));
+    const m = content.match(new RegExp(`(?:^|[\\s"'])${key}="([^"]+)"`));
     return m ? m[1] : null;
   };
   return {
@@ -23377,7 +23377,12 @@ function projectSubagentLine(line, agentId, state) {
       if (ct === "tool_use") {
         const name = c.name ?? "";
         if (name === "Agent" || name === "Task") {
-          events.push({ kind: "sub_agent_nested_spawn", agentId });
+          events.push({
+            kind: "sub_agent_nested_spawn",
+            agentId,
+            toolUseId: c.id ?? null,
+            input: c.input ?? undefined
+          });
         } else {
           events.push({
             kind: "sub_agent_tool_use",
@@ -24495,7 +24500,6 @@ function createIpcClient(options) {
 // bridge/tool-filter.ts
 var ALWAYS_LOAD_TOOLS = new Set([
   "reply",
-  "stream_reply",
   "get_recent_messages",
   "react",
   "edit_message",
@@ -24647,7 +24651,7 @@ var mcp = new Server({ name: "telegram", version: "1.0.0" }, {
   instructions: [
     "The sender reads Telegram, not this session. Anything you want them to see must go through the reply tool \u2014 your transcript output never reaches their chat.",
     "",
-    'Messages from Telegram arrive as <channel source="telegram" chat_id="..." message_id="..." user="..." ts="...">. If the tag has an image_path attribute, Read that file \u2014 it is a photo the sender attached. If the tag has attachment_file_id, call download_attachment with that file_id to fetch the file, then Read the returned path. A single message may carry SEVERAL attachments (a forwarded album or a text+multi-image burst): when attachment_count is set (>1), also handle the numbered siblings \u2014 image_path_2, image_path_3, \u2026 (Read each) and attachment_file_id_2, attachment_file_id_3, \u2026 (download_attachment each). Process every one, not just the first. Reply with the reply tool \u2014 pass chat_id back. The reply and stream_reply tools quote-reply to the latest inbound user message by default, so you do NOT need to pass reply_to for normal responses. Pass reply_to (a message_id) only when quoting a specific earlier message, or pass quote:false to send a bare (non-quoted) message.',
+    'Messages from Telegram arrive as <channel source="telegram" chat_id="..." message_id="..." user="..." ts="...">. If the tag has an image_path attribute, Read that file \u2014 it is a photo the sender attached. If the tag has attachment_file_id, call download_attachment with that file_id to fetch the file, then Read the returned path. A single message may carry SEVERAL attachments (a forwarded album or a text+multi-image burst): when attachment_count is set (>1), also handle the numbered siblings \u2014 image_path_2, image_path_3, \u2026 (Read each) and attachment_file_id_2, attachment_file_id_3, \u2026 (download_attachment each). Process every one, not just the first. Reply with the reply tool \u2014 pass chat_id back. The reply tool quote-replies to the latest inbound user message by default, so you do NOT need to pass reply_to for normal responses. Pass reply_to (a message_id) only when quoting a specific earlier message, or pass quote:false to send a bare (non-quoted) message.',
     "",
     `reply accepts file paths (files: ["/abs/path.png"]) for attachments. Use react to add emoji reactions, edit_message for interim progress updates, and delete_message when you need to truly remove a message (prefer edit_message if you just want to change text \u2014 delete is for retraction). Edits don't trigger push notifications \u2014 when a long task completes, send a new reply so the user's device pings. Use send_typing to show a typing indicator during long operations. Use pin_message to pin important outputs. Use forward_message to quote/resurface earlier messages.`,
     "",
@@ -24702,45 +24706,6 @@ var TOOL_SCHEMAS = [
       required: ["chat_id", "text"]
     }
   },
-  {
-    name: "stream_reply",
-    description: "Post the final answer for this turn. The plugin renders an event-driven progress card (Plan \u2192 Run \u2192 Done with live tool bullets, elapsed time, and status emoji) for free while the turn is in-flight, so you do not need to narrate intermediate progress. Call `stream_reply` exactly once per turn with done=true and the complete answer text. Hard cap is 32768 chars (the rich-message wire limit) \u2014 longer text is dropped by a defensive guard, so use `reply` for anything that long (it chunks). Calling with done=false is an error in this environment (the progress card already owns the mid-turn surface). inline_keyboard adds tappable buttons under the final message \u2014 see `reply` for shape and constraints.",
-    inputSchema: {
-      type: "object",
-      properties: {
-        chat_id: { type: "string" },
-        text: { type: "string", description: "Full text snapshot. NOT a delta \u2014 pass the complete current content each call." },
-        done: { type: "boolean", description: "Must be true. Posts this text as the final answer for the turn and locks the message." },
-        message_thread_id: { type: "string", description: "Forum topic thread ID. Auto-applied from the last inbound message if not specified." },
-        origin_turn_id: { type: "string", description: "In a forum supergroup, pass back the origin_turn_id attribute from the <channel> message you are answering. It pins the reply to that message's topic even if another topic's turn started meanwhile. Omit in DMs / single-topic chats." },
-        format: { type: "string", enum: ["html", "markdownv2", "text"], description: "Rendering mode. 'html' (default) converts markdown to Telegram HTML." },
-        reply_to: { type: "string", description: "Message ID to quote-reply to. Overrides the default (latest inbound)." },
-        quote: { type: "boolean", description: "Opt out of the default quote-reply behavior. Default: true. Ignored when reply_to is explicitly set." },
-        protect_content: { type: "boolean", description: "When true, Telegram prevents the message from being forwarded or saved." },
-        quote_text: { type: "string", description: "Surgical quote: specific text to highlight from the reply_to message. Requires reply_to." },
-        disable_notification: { type: "boolean", description: "When true, the INITIAL message send is silent (no device ping). Has no effect on subsequent edits \u2014 Telegram never pings on editMessageText. Default false. Use for mid-turn stream starts you do not want to ping; omit on the final answer." },
-        inline_keyboard: {
-          type: "array",
-          description: '2D array of tappable buttons under the final message. Same shape and constraints as `reply.inline_keyboard` \u2014 each button has `text` and EXACTLY ONE of `url` or `callback_data`, plus optional `ack_text` (custom tap-toast; default "\u2713 received") and `single_use` (default true; set false to keep the keyboard tappable after a tap). Tap on a callback_data button is delivered to this agent as an inbound channel event with meta.button_callback_data set.',
-          items: {
-            type: "array",
-            items: {
-              type: "object",
-              properties: {
-                text: { type: "string" },
-                url: { type: "string" },
-                callback_data: { type: "string" },
-                ack_text: { type: "string", description: 'Toast text shown on tap. Default "\u2713 received".' },
-                single_use: { type: "boolean", description: "Default true. Set false to keep the keyboard tappable after this button is tapped." }
-              },
-              required: ["text"]
-            }
-          }
-        }
-      },
-      required: ["chat_id", "text"]
-    }
-  },
   {
     name: "react",
     description: "Add an emoji reaction to a Telegram message. Telegram only accepts a fixed whitelist (\uD83D\uDC4D \uD83D\uDC4E \u2764 \uD83D\uDD25 \uD83D\uDC40 \uD83C\uDF89 etc) \u2014 non-whitelisted emoji will be rejected.",