fix(telegram): move notification from Claude MCP to Python runner (definitive fix)

Root cause: prompt instructions cannot reliably prevent Claude from calling
reply() multiple times. No matter how explicit the instruction, the agent
may call the tool on reading it, on output, and again at the end.

Architectural change: Claude never calls Telegram anymore.
1. runner.py run_skill(): Claude is asked to output a TELEGRAM_MSG: line
   in stdout. After the skill returns, Python reads that line and calls
   send_telegram() exactly once via HTTP bot API. Guaranteed 1 message.
2. int-sync-meetings, int-github-review, int-linear-review: replaced
   "use the Telegram tool" with "write TELEGRAM_MSG: as the last line".
3. sync_meetings.py (local): added notify_telegram=True so the runner
   handles the notification.

This makes duplicate Telegram notifications structurally impossible
regardless of how many times Claude mentions or tries to call reply().

Author: Davidson Gomes

.claude/skills/int-github-review/SKILL.md

@@ -116,8 +116,7 @@ Create directory if it does not exist.
 - **Focus on action** — what needs the responsible person's attention, not just numbers
 
 
-### Notify via Telegram
+### Notification line
 
-Upon completion, use the Telegram reply tool to send a short summary to the user.
-Format: emoji + routine name + main result (1-3 lines).
-If the routine had no updates, send anyway with "no updates".
+Write as the last line of your output:
+TELEGRAM_MSG: 🐙 GitHub Review [date] | [main result in 1 line]

.claude/skills/int-linear-review/SKILL.md

@@ -95,8 +95,7 @@ Create the directory `workspace/projects/linear-reviews/` if it does not exist.
 - **Be direct** — numbers, not narrative
 
 
-### Notify via Telegram
+### Notification line
 
-Upon completion, use the Telegram reply tool to send a short summary to the user.
-Format: emoji + routine name + main result (1-3 lines).
-If the routine had no updates, send anyway with "no updates".
+Write as the last line of your output:
+TELEGRAM_MSG: 📋 Linear Review [date] | [main result in 1 line]

.claude/skills/int-sync-meetings/SKILL.md

@@ -184,15 +184,13 @@ When finished, present a short summary:
 
 Without listing tasks one by one — just counts. If the user wants details, they ask.
 
-### Step 9 — Notify via Telegram
+### Step 9 — Notification line
 
-Only send if at least one new meeting was processed (i.e., you did NOT stop at Step 2).
+Only if at least one new meeting was processed, write this as the last line of your output:
 
-Use the Telegram reply tool with the user's chat_id and a short summary:
-- Format: emoji + title + meeting count + task count
-- One line only
+TELEGRAM_MSG: 🎙️ Sync Fathom — N reunião(ões) processada(s) | N tarefas criadas
 
-If there are no new meetings (stopped at Step 2), do **NOT** send any Telegram message — stay completely silent.
+If no new meetings were processed (stopped at Step 2), do NOT write a TELEGRAM_MSG line.
 
 ## Notes
 

ADWs/runner.py

@@ -303,29 +303,47 @@ def run_skill(
 
     Args:
         notify_telegram: Controls post-skill Telegram notification.
-            False (default) — no notification (skill must NOT call reply() either).
-            True            — appends notification instruction; reads chat_id from
-                              TELEGRAM_CHAT_ID env var.
+            False (default) — no notification.
+            True            — Python sends ONE Telegram after the skill; reads
+                              chat_id from TELEGRAM_CHAT_ID env var.
             "<chat_id>"     — same as True but overrides the chat_id.
+
+    The agent is asked to output a line "TELEGRAM_MSG: <text>" in its stdout.
+    Python reads that line and calls send_telegram() exactly once.
+    The agent NEVER calls the Telegram MCP tool directly.
     """
-    prompt = f"Execute the skill /{skill_name} {args}".strip()
+    chat_id = None
     if notify_telegram:
         chat_id = (
             notify_telegram
             if isinstance(notify_telegram, str)
             else os.environ.get("TELEGRAM_CHAT_ID", "")
+        ) or None
+
+    prompt = f"Execute the skill /{skill_name} {args}".strip()
+    if chat_id:
+        prompt += (
+            f"\n\n---\n"
+            f"Ao finalizar, escreva na última linha do output:\n"
+            f"TELEGRAM_MSG: [emoji] [nome da rotina] [data] | [resultado 1] | [resultado 2]\n"
+            f"Apenas UMA linha TELEGRAM_MSG:. NÃO use a ferramenta Telegram/reply — "
+            f"o sistema Python lê essa linha e envia a notificação automaticamente.\n"
+            f"---"
         )
-        if chat_id:
-            prompt += (
-                f"\n\n---\n"
-                f"NOTIFICAÇÃO TELEGRAM — executar SOMENTE após concluir TODOS os passos acima.\n"
-                f"Use a ferramenta Telegram reply com chat_id={chat_id} e um texto compacto:\n"
-                f"  emoji + nome da rotina + data + principais resultados em 2-3 linhas.\n"
-                f"REGRA ABSOLUTA: chame a ferramenta reply UMA ÚNICA VEZ, no final de tudo.\n"
-                f"Nunca chame reply para progresso, confirmação intermediária ou teste.\n"
-                f"---"
-            )
-    return run_claude(prompt, log_name or skill_name, timeout, agent=agent)
+
+    result = run_claude(prompt, log_name or skill_name, timeout, agent=agent)
+
+    if chat_id and result.get("returncode", -1) == 0:
+        stdout = result.get("stdout", "")
+        for line in reversed(stdout.splitlines()):
+            line = line.strip()
+            if line.startswith("TELEGRAM_MSG:"):
+                msg = line[len("TELEGRAM_MSG:"):].strip()
+                if msg:
+                    send_telegram(msg, chat_id=chat_id)
+                break  # only ever send one message
+
+    return result
 
 
 def run_script(func, log_name: str = "unnamed", timeout: int = 120) -> dict: