hcom 0.1.4__tar.gz → 0.1.6__tar.gz

{hcom-0.1.4/src/hcom.egg-info → hcom-0.1.6}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: hcom
-Version: 0.1.4
-Summary: Lightweight CLI tool for real-time messaging between Claude Code subagents using hooks
+Version: 0.1.6
+Summary: Lightweight CLI tool for real-time communication between Claude Code subagents using hooks
 Author-email: aannoo <your@email.com>
 License: MIT
 Project-URL: Homepage, https://github.com/aannoo/claude-hook-comms
@@ -26,9 +26,9 @@ Classifier: Topic :: Communications
 Requires-Python: >=3.6
 Description-Content-Type: text/markdown
 
-# Claude Code Hook Comms
+# hcom - Claude Hook Comms
 
-Lightweight CLI tool for real-time communication between claude code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
+Lightweight CLI tool for real-time communication between Claude Code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
 
 ## 🦆 What It Does
 
@@ -38,7 +38,7 @@ Creates a group chat where you and multiple interactive Claude Code subagents ca
 
 ## 🦷 Features
 
-- **Multi-Terminal Launch** - Launch claude code subagents in new terminals
+- **Multi-Terminal Launch** - Launch Claude Code subagents in new terminals
 - **Live Dashboard** - Real-time monitoring of all instances
 - **Multi-Agent Communication** - Claude instances talk to each other across projects
 - **@Mention Targeting** - Send messages to specific subagents or teams
@@ -154,7 +154,7 @@ HCOM_INSTANCE_HINTS="always update chat with progress" hcom open nice-subagent-b
 | `max_messages_per_delivery` | 50 | `HCOM_MAX_MESSAGES_PER_DELIVERY` | Messages delivered per batch |
 | `sender_name` | "bigboss" | `HCOM_SENDER_NAME` | Your name in chat |
 | `sender_emoji` | "🐳" | `HCOM_SENDER_EMOJI` | Your emoji icon |
-| `initial_prompt` | "Say hi" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
+| `initial_prompt` | "Say hi in chat" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
 | `first_use_text` | "Essential, concise messages only" | `HCOM_FIRST_USE_TEXT` | Welcome message for instances |
 | `terminal_mode` | "new_window" | `HCOM_TERMINAL_MODE` | How to launch terminals ("new_window", "same_terminal", "show_commands") |
 | `terminal_command` | null | `HCOM_TERMINAL_COMMAND` | Custom terminal command (see Terminal Options) |
@@ -202,11 +202,11 @@ hcom adds hooks to your project directory's `.claude/settings.local.json`:
 - **Identity**: Each instance gets a unique name based on conversation UUID (e.g., "hovoa7")
 - **Persistence**: Names persist across `--resume` maintaining conversation context
 - **Status Detection**: Notification hook tracks permission requests and activity
-- **Agents**: When you run `hcom open researcher`, it loads an interactive claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global)
+- **Agents**: When you run `hcom open researcher`, it loads an interactive Claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global). Agents can specify `model:` and `tools:` in YAML frontmatter
 
 ### Architecture
 - **Single conversation** - All instances share one global conversation
-- **Opt-in participation** - Only claude code instances launched with `hcom open` join the chat
+- **Opt-in participation** - Only Claude Code instances launched with `hcom open` join the chat
 - **@-mention filtering** - Target messages to specific instances or teams
 
 ### File Structure

{hcom-0.1.4 → hcom-0.1.6}/README.md

@@ -1,6 +1,6 @@
-# Claude Code Hook Comms
+# hcom - Claude Hook Comms
 
-Lightweight CLI tool for real-time communication between claude code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
+Lightweight CLI tool for real-time communication between Claude Code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
 
 ## 🦆 What It Does
 
@@ -10,7 +10,7 @@ Creates a group chat where you and multiple interactive Claude Code subagents ca
 
 ## 🦷 Features
 
-- **Multi-Terminal Launch** - Launch claude code subagents in new terminals
+- **Multi-Terminal Launch** - Launch Claude Code subagents in new terminals
 - **Live Dashboard** - Real-time monitoring of all instances
 - **Multi-Agent Communication** - Claude instances talk to each other across projects
 - **@Mention Targeting** - Send messages to specific subagents or teams
@@ -126,7 +126,7 @@ HCOM_INSTANCE_HINTS="always update chat with progress" hcom open nice-subagent-b
 | `max_messages_per_delivery` | 50 | `HCOM_MAX_MESSAGES_PER_DELIVERY` | Messages delivered per batch |
 | `sender_name` | "bigboss" | `HCOM_SENDER_NAME` | Your name in chat |
 | `sender_emoji` | "🐳" | `HCOM_SENDER_EMOJI` | Your emoji icon |
-| `initial_prompt` | "Say hi" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
+| `initial_prompt` | "Say hi in chat" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
 | `first_use_text` | "Essential, concise messages only" | `HCOM_FIRST_USE_TEXT` | Welcome message for instances |
 | `terminal_mode` | "new_window" | `HCOM_TERMINAL_MODE` | How to launch terminals ("new_window", "same_terminal", "show_commands") |
 | `terminal_command` | null | `HCOM_TERMINAL_COMMAND` | Custom terminal command (see Terminal Options) |
@@ -174,11 +174,11 @@ hcom adds hooks to your project directory's `.claude/settings.local.json`:
 - **Identity**: Each instance gets a unique name based on conversation UUID (e.g., "hovoa7")
 - **Persistence**: Names persist across `--resume` maintaining conversation context
 - **Status Detection**: Notification hook tracks permission requests and activity
-- **Agents**: When you run `hcom open researcher`, it loads an interactive claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global)
+- **Agents**: When you run `hcom open researcher`, it loads an interactive Claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global). Agents can specify `model:` and `tools:` in YAML frontmatter
 
 ### Architecture
 - **Single conversation** - All instances share one global conversation
-- **Opt-in participation** - Only claude code instances launched with `hcom open` join the chat
+- **Opt-in participation** - Only Claude Code instances launched with `hcom open` join the chat
 - **@-mention filtering** - Target messages to specific instances or teams
 
 ### File Structure

{hcom-0.1.4 → hcom-0.1.6}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "hcom"
-version = "0.1.4"
-description = "Lightweight CLI tool for real-time messaging between Claude Code subagents using hooks"
+version = "0.1.6"
+description = "Lightweight CLI tool for real-time communication between Claude Code subagents using hooks"
 readme = "README.md"
 requires-python = ">=3.6"
 license = {text = "MIT"}

{hcom-0.1.4 → hcom-0.1.6}/src/hcom/__init__.py

@@ -1,3 +1,3 @@
 """Claude Hook Comms - Real-time messaging between Claude Code agents."""
 
-__version__ = "0.1.3"
+__version__ = "0.1.5"

{hcom-0.1.4 → hcom-0.1.6}/src/hcom/__main__.py

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 """
 hcom - Claude Hook Comms
-A lightweight multi-agent communication system for claude code
+Lightweight CLI tool for real-time communication between Claude Code subagents using hooks
 """
 
 import os
@@ -170,27 +170,22 @@ def get_config_value(key, default=None):
     return config.get(key, default)
 
 def get_hook_command():
-    """Determine the best hook command approach based on paths"""
+    """Get hook command with silent fallback
+    
+    Uses ${HCOM:-true} for clean paths, conditional for paths with spaces.
+    Both approaches exit silently (code 0) when not launched via 'hcom open'.
+    """
     python_path = sys.executable
     script_path = os.path.abspath(__file__)
     
-    # Characters that cause issues in shell expansion
-    problematic_chars = [' ', '`', '"', "'", '\\', '\n', ';', '&', '|', '(', ')', ':', 
-                        '$', '*', '?', '[', ']', '{', '}', '!', '~', '<', '>']
-    
-    has_problematic_chars = any(char in python_path or char in script_path 
-                               for char in problematic_chars)
-    
-    if has_problematic_chars:
-        # Use direct paths with proper escaping
-        # Escape backslashes first, then quotes
-        escaped_python = python_path.replace('\\', '\\\\').replace('"', '\\"')
-        escaped_script = script_path.replace('\\', '\\\\').replace('"', '\\"')
-        return f'"{escaped_python}" "{escaped_script}"', {}
+    if ' ' in python_path or ' ' in script_path:
+        # Paths with spaces: use conditional check
+        escaped_python = shlex.quote(python_path)
+        escaped_script = shlex.quote(script_path)
+        return f'[ "${{HCOM_ACTIVE}}" = "1" ] && {escaped_python} {escaped_script} || true', {}
     else:
-        # Use single clean env var
-        env_vars = {'HCOM': f'{python_path} {script_path}'}
-        return '$HCOM', env_vars
+        # Clean paths: use environment variable
+        return '${HCOM:-true}', {}
 
 def build_claude_env():
     """Build environment variables for Claude instances"""
@@ -206,9 +201,11 @@ def build_claude_env():
     
     env.update(config.get('env_overrides', {}))
     
-    # Add hook-specific env vars if using that approach
-    _, hook_env_vars = get_hook_command()
-    env.update(hook_env_vars)
+    # Set HCOM only for clean paths (spaces handled differently)
+    python_path = sys.executable
+    script_path = os.path.abspath(__file__)
+    if ' ' not in python_path and ' ' not in script_path:
+        env['HCOM'] = f'{python_path} {script_path}'
     
     return env
 
@@ -335,6 +332,35 @@ def parse_open_args(args):
     
     return instances, prefix, claude_args
 
+def extract_agent_config(content):
+    """Extract configuration from agent YAML frontmatter"""
+    if not content.startswith('---'):
+        return {}
+    
+    # Find YAML section between --- markers
+    yaml_end = content.find('\n---', 3)
+    if yaml_end < 0:
+        return {}  # No closing marker
+    
+    yaml_section = content[3:yaml_end]
+    config = {}
+    
+    # Extract model field
+    model_match = re.search(r'^model:\s*(.+)$', yaml_section, re.MULTILINE)
+    if model_match:
+        value = model_match.group(1).strip()
+        if value and value.lower() != 'inherit':
+            config['model'] = value
+    
+    # Extract tools field
+    tools_match = re.search(r'^tools:\s*(.+)$', yaml_section, re.MULTILINE)
+    if tools_match:
+        value = tools_match.group(1).strip()
+        if value:
+            config['tools'] = value.replace(', ', ',')
+    
+    return config
+
 def resolve_agent(name):
     """Resolve agent file by name
     
@@ -342,16 +368,17 @@ def resolve_agent(name):
     1. .claude/agents/{name}.md (local)
     2. ~/.claude/agents/{name}.md (global)
     
-    Returns the content after stripping YAML frontmatter
+    Returns tuple: (content after stripping YAML frontmatter, config dict)
     """
     for base_path in [Path('.'), Path.home()]:
         agent_path = base_path / '.claude/agents' / f'{name}.md'
         if agent_path.exists():
             content = agent_path.read_text()
+            config = extract_agent_config(content)
             stripped = strip_frontmatter(content)
             if not stripped.strip():
                 raise ValueError(format_error(f"Agent '{name}' has empty content", 'Check the agent file contains a system prompt'))
-            return stripped
+            return stripped, config
     
     raise FileNotFoundError(format_error(f'Agent not found: {name}', 'Check available agents or create the agent file'))
 
@@ -393,15 +420,22 @@ def _remove_hcom_hooks_from_settings(settings):
     
     # Patterns to match any hcom hook command
     # - $HCOM post/stop/notify
+    # - ${HCOM:-...} post/stop/notify
+    # - [ "${HCOM_ACTIVE}" = "1" ] && ... hcom.py ... || true
     # - hcom post/stop/notify  
+    # - uvx hcom post/stop/notify
     # - /path/to/hcom.py post/stop/notify
+    # - sh -c "[ ... ] && ... hcom ..."
     # - "/path with spaces/python" "/path with spaces/hcom.py" post/stop/notify
     # - '/path/to/python' '/path/to/hcom.py' post/stop/notify
     hcom_patterns = [
-        r'\$HCOM\s+(post|stop|notify)\b',           # Environment variable
+        r'\$\{?HCOM',                                # Environment variable (with or without braces)
+        r'\bHCOM_ACTIVE.*hcom\.py',                 # Conditional with HCOM_ACTIVE check
         r'\bhcom\s+(post|stop|notify)\b',           # Direct hcom command
+        r'\buvx\s+hcom\s+(post|stop|notify)\b',     # uvx hcom command
         r'hcom\.py["\']?\s+(post|stop|notify)\b',   # hcom.py with optional quote
         r'["\'][^"\']*hcom\.py["\']?\s+(post|stop|notify)\b',  # Quoted path with hcom.py
+        r'sh\s+-c.*hcom',                           # Shell wrapper with hcom
     ]
     compiled_patterns = [re.compile(pattern) for pattern in hcom_patterns]
     
@@ -452,7 +486,7 @@ def format_warning(message):
     """Format warning message consistently"""
     return f"Warning: {message}"
 
-def build_claude_command(agent_content=None, claude_args=None, initial_prompt="Say hi in chat"):
+def build_claude_command(agent_content=None, claude_args=None, initial_prompt="Say hi in chat", model=None, tools=None):
     """Build Claude command with proper argument handling
     
     Returns tuple: (command_string, temp_file_path_or_none)
@@ -461,6 +495,27 @@ def build_claude_command(agent_content=None, claude_args=None, initial_prompt="S
     cmd_parts = ['claude']
     temp_file_path = None
     
+    # Add model if specified and not already in claude_args
+    if model:
+        # Check if model already specified in args (more concise)
+        has_model = claude_args and any(
+            arg in ['--model', '-m'] or 
+            arg.startswith(('--model=', '-m=')) 
+            for arg in claude_args
+        )
+        if not has_model:
+            cmd_parts.extend(['--model', model])
+    
+    # Add allowed tools if specified and not already in claude_args
+    if tools:
+        has_tools = claude_args and any(
+            arg in ['--allowedTools', '--allowed-tools'] or
+            arg.startswith(('--allowedTools=', '--allowed-tools='))
+            for arg in claude_args
+        )
+        if not has_tools:
+            cmd_parts.extend(['--allowedTools', tools])
+    
     if claude_args:
         for arg in claude_args:
             cmd_parts.append(shlex.quote(arg))
@@ -662,7 +717,7 @@ def setup_hooks():
     if hcom_send_permission not in settings['permissions']['allow']:
         settings['permissions']['allow'].append(hcom_send_permission)
     
-    # Get the hook command approach (env vars or direct paths based on spaces)
+    # Get the hook command template
     hook_cmd_base, _ = get_hook_command()
     
     # Add PostToolUse hook
@@ -718,16 +773,36 @@ def get_archive_timestamp():
     return datetime.now().strftime("%Y%m%d-%H%M%S")
 
 def get_conversation_uuid(transcript_path):
-    """Get conversation UUID from transcript"""
+    """Get conversation UUID from transcript
+    
+    For resumed sessions, the first line may be a summary with a different leafUuid.
+    We need to find the first user entry which contains the stable conversation UUID.
+    """
     try:
         if not transcript_path or not os.path.exists(transcript_path):
             return None
         
+        # First, try to find the UUID from the first user entry
+        with open(transcript_path, 'r') as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    entry = json.loads(line)
+                    # Look for first user entry with a UUID - this is the stable identifier
+                    if entry.get('type') == 'user' and entry.get('uuid'):
+                        return entry.get('uuid')
+                except json.JSONDecodeError:
+                    continue
+        
+        # Fallback: If no user entry found, try the first line (original behavior)
         with open(transcript_path, 'r') as f:
             first_line = f.readline().strip()
             if first_line:
                 entry = json.loads(first_line)
-                return entry.get('uuid')
+                # Try both 'uuid' and 'leafUuid' fields
+                return entry.get('uuid') or entry.get('leafUuid')
     except Exception:
         pass
     return None
@@ -1119,7 +1194,8 @@ def migrate_instance_name_if_needed(instance_name, conversation_uuid, transcript
     if instance_name.endswith("claude") and conversation_uuid:
         new_instance = get_display_name(transcript_path)
         if new_instance != instance_name and not new_instance.endswith("claude"):
-            # Migrate from fallback name to UUID-based name
+            # Always return the new name if we can generate it
+            # Migration of data only happens if old name exists
             pos_file = get_hcom_dir() / "hcom.json"
             positions = load_positions(pos_file)
             if instance_name in positions:
@@ -1128,8 +1204,7 @@ def migrate_instance_name_if_needed(instance_name, conversation_uuid, transcript
                 # Update the conversation UUID in the migrated data
                 positions[new_instance]["conversation_uuid"] = conversation_uuid
                 atomic_write(pos_file, json.dumps(positions, indent=2))
-                # Instance name migrated
-                return new_instance
+            return new_instance
     return instance_name
 
 def update_instance_position(instance_name, update_fields):
@@ -1282,11 +1357,16 @@ def cmd_open(*args):
             else:
                 # Agent instance
                 try:
-                    agent_content = resolve_agent(instance_type)
+                    agent_content, agent_config = resolve_agent(instance_type)
+                    # Use agent's model and tools if specified and not overridden in claude_args
+                    agent_model = agent_config.get('model')
+                    agent_tools = agent_config.get('tools')
                     claude_cmd, temp_file = build_claude_command(
                         agent_content=agent_content,
                         claude_args=claude_args,
-                        initial_prompt=initial_prompt
+                        initial_prompt=initial_prompt,
+                        model=agent_model,
+                        tools=agent_tools
                     )
                     if temp_file:
                         temp_files_to_cleanup.append(temp_file)
@@ -1746,12 +1826,9 @@ def handle_hook_post():
         # Migrate instance name if needed (from fallback to UUID-based)
         instance_name = migrate_instance_name_if_needed(instance_name, conversation_uuid, transcript_path)
         
-        # Initialize instance if needed
-        if not instance_name.endswith("claude") or conversation_uuid:
-            initialize_instance_in_position_file(instance_name, conversation_uuid)
-            
-            # Update instance position
-            update_instance_position(instance_name, {
+        initialize_instance_in_position_file(instance_name, conversation_uuid)
+        
+        update_instance_position(instance_name, {
                 'last_tool': int(time.time()),
                 'last_tool_name': hook_data.get('tool_name', 'unknown'),
                 'session_id': hook_data.get('session_id', ''),
@@ -1788,8 +1865,7 @@ def handle_hook_post():
                     elif message and message[-1] in '"\'':
                         message = message[:-1]
                     
-                    if message and not instance_name.endswith("claude"):
-                        # Validate message
+                    if message:
                         error = validate_message(message)
                         if error:
                             output = {"reason": f"❌ {error}"}
@@ -1799,27 +1875,25 @@ def handle_hook_post():
                         send_message(instance_name, message)
                         sent_reason = "✓ Sent"
         
-        # Check for pending messages to deliver
-        if not instance_name.endswith("claude"):
-            messages = get_new_messages(instance_name)
-            
-            if messages and sent_reason:
-                # Both sent and received
-                reason = f"{sent_reason} | {format_hook_messages(messages, instance_name)}"
-                output = {"decision": HOOK_DECISION_BLOCK, "reason": reason}
-                print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
-                sys.exit(EXIT_BLOCK)
-            elif messages:
-                # Just received
-                reason = format_hook_messages(messages, instance_name)
-                output = {"decision": HOOK_DECISION_BLOCK, "reason": reason}
-                print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
-                sys.exit(EXIT_BLOCK)
-            elif sent_reason:
-                # Just sent
-                output = {"reason": sent_reason}
-                print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
-                sys.exit(EXIT_BLOCK)
+        messages = get_new_messages(instance_name)
+        
+        if messages and sent_reason:
+            # Both sent and received
+            reason = f"{sent_reason} | {format_hook_messages(messages, instance_name)}"
+            output = {"decision": HOOK_DECISION_BLOCK, "reason": reason}
+            print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
+            sys.exit(EXIT_BLOCK)
+        elif messages:
+            # Just received
+            reason = format_hook_messages(messages, instance_name)
+            output = {"decision": HOOK_DECISION_BLOCK, "reason": reason}
+            print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
+            sys.exit(EXIT_BLOCK)
+        elif sent_reason:
+            # Just sent
+            output = {"reason": sent_reason}
+            print(json.dumps(output, ensure_ascii=False), file=sys.stderr)
+            sys.exit(EXIT_BLOCK)
     
     except Exception:
         pass
@@ -1839,9 +1913,6 @@ def handle_hook_stop():
         instance_name = get_display_name(transcript_path) if transcript_path else f"{Path.cwd().name[:2].lower()}claude"
         conversation_uuid = get_conversation_uuid(transcript_path)
         
-        if instance_name.endswith("claude") and not conversation_uuid:
-            sys.exit(EXIT_SUCCESS)
-        
         # Initialize instance if needed
         initialize_instance_in_position_file(instance_name, conversation_uuid)
         
@@ -1906,9 +1977,6 @@ def handle_hook_notification():
         instance_name = get_display_name(transcript_path) if transcript_path else f"{Path.cwd().name[:2].lower()}claude"
         conversation_uuid = get_conversation_uuid(transcript_path)
         
-        if instance_name.endswith("claude") and not conversation_uuid:
-            sys.exit(EXIT_SUCCESS)
-        
         # Initialize instance if needed
         initialize_instance_in_position_file(instance_name, conversation_uuid)
         

{hcom-0.1.4 → hcom-0.1.6/src/hcom.egg-info}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: hcom
-Version: 0.1.4
-Summary: Lightweight CLI tool for real-time messaging between Claude Code subagents using hooks
+Version: 0.1.6
+Summary: Lightweight CLI tool for real-time communication between Claude Code subagents using hooks
 Author-email: aannoo <your@email.com>
 License: MIT
 Project-URL: Homepage, https://github.com/aannoo/claude-hook-comms
@@ -26,9 +26,9 @@ Classifier: Topic :: Communications
 Requires-Python: >=3.6
 Description-Content-Type: text/markdown
 
-# Claude Code Hook Comms
+# hcom - Claude Hook Comms
 
-Lightweight CLI tool for real-time communication between claude code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
+Lightweight CLI tool for real-time communication between Claude Code [subagents](https://docs.anthropic.com/en/docs/claude-code/sub-agents) using [hooks](https://docs.anthropic.com/en/docs/claude-code/hooks).
 
 ## 🦆 What It Does
 
@@ -38,7 +38,7 @@ Creates a group chat where you and multiple interactive Claude Code subagents ca
 
 ## 🦷 Features
 
-- **Multi-Terminal Launch** - Launch claude code subagents in new terminals
+- **Multi-Terminal Launch** - Launch Claude Code subagents in new terminals
 - **Live Dashboard** - Real-time monitoring of all instances
 - **Multi-Agent Communication** - Claude instances talk to each other across projects
 - **@Mention Targeting** - Send messages to specific subagents or teams
@@ -154,7 +154,7 @@ HCOM_INSTANCE_HINTS="always update chat with progress" hcom open nice-subagent-b
 | `max_messages_per_delivery` | 50 | `HCOM_MAX_MESSAGES_PER_DELIVERY` | Messages delivered per batch |
 | `sender_name` | "bigboss" | `HCOM_SENDER_NAME` | Your name in chat |
 | `sender_emoji` | "🐳" | `HCOM_SENDER_EMOJI` | Your emoji icon |
-| `initial_prompt` | "Say hi" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
+| `initial_prompt` | "Say hi in chat" | `HCOM_INITIAL_PROMPT` | What new instances are told to do |
 | `first_use_text` | "Essential, concise messages only" | `HCOM_FIRST_USE_TEXT` | Welcome message for instances |
 | `terminal_mode` | "new_window" | `HCOM_TERMINAL_MODE` | How to launch terminals ("new_window", "same_terminal", "show_commands") |
 | `terminal_command` | null | `HCOM_TERMINAL_COMMAND` | Custom terminal command (see Terminal Options) |
@@ -202,11 +202,11 @@ hcom adds hooks to your project directory's `.claude/settings.local.json`:
 - **Identity**: Each instance gets a unique name based on conversation UUID (e.g., "hovoa7")
 - **Persistence**: Names persist across `--resume` maintaining conversation context
 - **Status Detection**: Notification hook tracks permission requests and activity
-- **Agents**: When you run `hcom open researcher`, it loads an interactive claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global)
+- **Agents**: When you run `hcom open researcher`, it loads an interactive Claude session with a system prompt from `.claude/agents/researcher.md` (local) or `~/.claude/agents/researcher.md` (global). Agents can specify `model:` and `tools:` in YAML frontmatter
 
 ### Architecture
 - **Single conversation** - All instances share one global conversation
-- **Opt-in participation** - Only claude code instances launched with `hcom open` join the chat
+- **Opt-in participation** - Only Claude Code instances launched with `hcom open` join the chat
 - **@-mention filtering** - Target messages to specific instances or teams
 
 ### File Structure