aru-code 0.10.0__tar.gz → 0.11.0__tar.gz

{aru_code-0.10.0/aru_code.egg-info → aru_code-0.11.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aru-code
-Version: 0.10.0
+Version: 0.11.0
 Summary: A Claude Code clone built with Agno agents
 Author-email: Estevao <estevaofon@gmail.com>
 License-Expression: MIT
@@ -55,6 +55,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 
 - **Multi-Agent Architecture** — Specialized agents for planning, execution, and conversation
 - **Interactive CLI** — Streaming responses, multi-line paste, session management
+- **Image Support** — Attach images via `@` mentions for multimodal analysis (Claude, GPT-4o, Gemini)
 - **16 Integrated Tools** — File operations, code search, shell, web search, task delegation
 - **Task Planning** — Break down complex tasks into steps with automatic execution
 - **Multi-Provider** — Anthropic, OpenAI, Ollama, Groq, OpenRouter, DeepSeek, and others via custom configuration
@@ -69,7 +70,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 pip install aru-code
 ```
 
-> **Requirements:** Python 3.13+
+> **Requirements:** Python 3.11+
 
 ### 2. Configure the API Key
 
@@ -131,6 +132,23 @@ aru> ! pytest tests/ -v
 aru> /model ollama/codellama
 ```
 
+### Image Support
+
+Attach images to your messages using the same `@` mention syntax used for files. Aru detects image files by extension and sends them to the LLM as visual content for multimodal analysis.
+
+**Supported formats:** `.png`, `.jpg`, `.jpeg`, `.gif`, `.webp`, `.bmp`
+
+```
+aru> describe @screenshot.png
+aru> compare @before.png and @after.png
+aru> review @code.py and explain the diagram in @architecture.png
+aru> analyze @D:/full/path/to/image.jpg
+```
+
+Images are sent natively to the model via the provider's multimodal API — no base64 text is injected into the conversation. Works with any multimodal model (Claude Opus/Sonnet, GPT-4o, Gemini, etc.). The autocomplete shows an `[image]` label for image files.
+
+> **Note:** Images require a multimodal model. Local models via Ollama may not support image input. Maximum file size: 20MB.
+
 ## Configuration
 
 ### Models and Providers
@@ -374,11 +392,22 @@ performance, and readability. Do NOT modify files.
 
 #### Invocation
 
+There are three ways to invoke a custom agent:
+
+| Method | Syntax | When to use |
+|--------|--------|-------------|
+| **Slash command** | `/reviewer src/auth.py` | Directly invoke a `primary` agent by name |
+| **@mention** | `@reviewer check this function` | Mention an agent anywhere in your message |
+| **delegate_task** | Automatic (subagents only) | Subagent names and descriptions are injected into the `delegate_task` tool description, so the LLM sees them and can call `delegate_task(task="...", agent="name")` on its own when it judges the task fits |
+
 ```
-aru> /reviewer src/auth.py        # invoke by slash + filename (without .md)
-aru> /agents                       # list all custom agents
+aru> /reviewer src/auth.py           # slash command (primary agents)
+aru> @reviewer check the auth module  # @mention (primary or subagent)
+aru> /agents                          # list all custom agents
 ```
 
+> **Note:** Slash commands (`/name`) are only available for `primary` agents — subagents are blocked with a warning. `@mention` works for any agent regardless of mode. Subagents can be invoked in two ways: automatically by the LLM via `delegate_task`, or manually by the user via `@name`.
+
 #### Discovery paths
 
 Agents are discovered from multiple locations (later overrides earlier):

{aru_code-0.10.0 → aru_code-0.11.0}/README.md

@@ -8,6 +8,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 
 - **Multi-Agent Architecture** — Specialized agents for planning, execution, and conversation
 - **Interactive CLI** — Streaming responses, multi-line paste, session management
+- **Image Support** — Attach images via `@` mentions for multimodal analysis (Claude, GPT-4o, Gemini)
 - **16 Integrated Tools** — File operations, code search, shell, web search, task delegation
 - **Task Planning** — Break down complex tasks into steps with automatic execution
 - **Multi-Provider** — Anthropic, OpenAI, Ollama, Groq, OpenRouter, DeepSeek, and others via custom configuration
@@ -22,7 +23,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 pip install aru-code
 ```
 
-> **Requirements:** Python 3.13+
+> **Requirements:** Python 3.11+
 
 ### 2. Configure the API Key
 
@@ -84,6 +85,23 @@ aru> ! pytest tests/ -v
 aru> /model ollama/codellama
 ```
 
+### Image Support
+
+Attach images to your messages using the same `@` mention syntax used for files. Aru detects image files by extension and sends them to the LLM as visual content for multimodal analysis.
+
+**Supported formats:** `.png`, `.jpg`, `.jpeg`, `.gif`, `.webp`, `.bmp`
+
+```
+aru> describe @screenshot.png
+aru> compare @before.png and @after.png
+aru> review @code.py and explain the diagram in @architecture.png
+aru> analyze @D:/full/path/to/image.jpg
+```
+
+Images are sent natively to the model via the provider's multimodal API — no base64 text is injected into the conversation. Works with any multimodal model (Claude Opus/Sonnet, GPT-4o, Gemini, etc.). The autocomplete shows an `[image]` label for image files.
+
+> **Note:** Images require a multimodal model. Local models via Ollama may not support image input. Maximum file size: 20MB.
+
 ## Configuration
 
 ### Models and Providers
@@ -327,11 +345,22 @@ performance, and readability. Do NOT modify files.
 
 #### Invocation
 
+There are three ways to invoke a custom agent:
+
+| Method | Syntax | When to use |
+|--------|--------|-------------|
+| **Slash command** | `/reviewer src/auth.py` | Directly invoke a `primary` agent by name |
+| **@mention** | `@reviewer check this function` | Mention an agent anywhere in your message |
+| **delegate_task** | Automatic (subagents only) | Subagent names and descriptions are injected into the `delegate_task` tool description, so the LLM sees them and can call `delegate_task(task="...", agent="name")` on its own when it judges the task fits |
+
 ```
-aru> /reviewer src/auth.py        # invoke by slash + filename (without .md)
-aru> /agents                       # list all custom agents
+aru> /reviewer src/auth.py           # slash command (primary agents)
+aru> @reviewer check the auth module  # @mention (primary or subagent)
+aru> /agents                          # list all custom agents
 ```
 
+> **Note:** Slash commands (`/name`) are only available for `primary` agents — subagents are blocked with a warning. `@mention` works for any agent regardless of mode. Subagents can be invoked in two ways: automatically by the LLM via `delegate_task`, or manually by the user via `@name`.
+
 #### Discovery paths
 
 Agents are discovered from multiple locations (later overrides earlier):

aru_code-0.11.0/aru/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.11.0"

{aru_code-0.10.0 → aru_code-0.11.0}/aru/cli.py

@@ -230,9 +230,15 @@ async def run_cli(skip_permissions: bool = False, resume_id: str | None = None):
 
         # Resolve @file mentions (skip known agent names)
         _agent_names = set(config.custom_agents.keys()) if config.custom_agents else set()
-        resolved, injected = _resolve_mentions(user_input, os.getcwd(), _agent_names)
-        if resolved != user_input:
-            console.print(f"[dim]Attached {injected} file(s) from @ mentions[/dim]")
+        resolved, injected, attached_images = _resolve_mentions(user_input, os.getcwd(), _agent_names)
+        if injected > 0:
+            parts = []
+            text_count = injected - len(attached_images)
+            if text_count > 0:
+                parts.append(f"{text_count} file(s)")
+            if attached_images:
+                parts.append(f"{len(attached_images)} image(s)")
+            console.print(f"[dim]Attached {', '.join(parts)} from @ mentions[/dim]")
             user_input = resolved
 
         if paste_state.pasted_content and user_text:
@@ -441,7 +447,7 @@ async def run_cli(skip_permissions: bool = False, resume_id: str | None = None):
                 else:
                     agent = create_general_agent(session, config)
                 session.add_message("user", user_input)
-                run_result = await run_agent_capture(agent, prompt, session)
+                run_result = await run_agent_capture(agent, prompt, session, images=attached_images or None)
                 if run_result.content:
                     session.add_message("assistant", run_result.with_tools_summary())
             elif cmd_name in config.skills:
@@ -454,7 +460,7 @@ async def run_cli(skip_permissions: bool = False, resume_id: str | None = None):
 
                     agent = create_general_agent(session, config)
                     session.add_message("user", user_input)
-                    run_result = await run_agent_capture(agent, prompt, session)
+                    run_result = await run_agent_capture(agent, prompt, session, images=attached_images or None)
                     if run_result.content:
                         session.add_message("assistant", run_result.with_tools_summary())
             elif cmd_name in config.custom_agents:
@@ -467,7 +473,7 @@ async def run_cli(skip_permissions: bool = False, resume_id: str | None = None):
                     agent = create_custom_agent_instance(agent_def, session, config)
                     session.add_message("user", user_input)
                     with permission_scope(agent_def.permission):
-                        run_result = await run_agent_capture(agent, cmd_args or user_input, session)
+                        run_result = await run_agent_capture(agent, cmd_args or user_input, session, images=attached_images or None)
                     if run_result.content:
                         session.add_message("assistant", run_result.with_tools_summary())
             else:
@@ -495,13 +501,13 @@ async def run_cli(skip_permissions: bool = False, resume_id: str | None = None):
                 agent = create_custom_agent_instance(agent_def, session, config)
                 session.add_message("user", user_input)
                 with permission_scope(agent_def.permission):
-                    run_result = await run_agent_capture(agent, message_text, session)
+                    run_result = await run_agent_capture(agent, message_text, session, images=attached_images or None)
                 if run_result.content:
                     session.add_message("assistant", run_result.with_tools_summary())
             else:
                 agent = create_general_agent(session, config)
                 session.add_message("user", user_input)
-                run_result = await run_agent_capture(agent, user_input, session)
+                run_result = await run_agent_capture(agent, user_input, session, images=attached_images or None)
                 if run_result.content:
                     session.add_message("assistant", run_result.with_tools_summary())
 

{aru_code-0.10.0 → aru_code-0.11.0}/aru/completers.py

@@ -12,25 +12,31 @@ from prompt_toolkit.formatted_text import HTML
 from prompt_toolkit.key_binding import KeyBindings
 from prompt_toolkit.keys import Keys
 
+from agno.media import Image
+
 from aru.commands import SLASH_COMMANDS
 from aru.config import AgentConfig
 
-_MENTION_RE = re.compile(r'(?<!\S)@([a-zA-Z0-9_./\\-]+)')
+_MENTION_RE = re.compile(r'(?<!\S)@([a-zA-Z0-9_./\\:-]+)')
 _MENTION_MAX_SIZE = 30_000  # bytes, same limit as read_file
+_IMAGE_EXTENSIONS = {".png", ".jpg", ".jpeg", ".gif", ".webp", ".bmp"}
+_IMAGE_MAX_SIZE = 20 * 1024 * 1024  # 20MB
 
 
-def _resolve_mentions(text: str, cwd: str, agent_names: set[str] | None = None) -> tuple[str, int]:
+def _resolve_mentions(text: str, cwd: str, agent_names: set[str] | None = None) -> tuple[str, int, list[Image]]:
     """Resolve @file mentions by appending file contents to the message.
 
+    Image files (png, jpg, etc.) are returned as Image objects instead of text.
     Skips @mentions that match known agent names.
-    Returns (resolved_text, number_of_files_attached).
+    Returns (resolved_text, number_of_files_attached, images).
     """
     agent_names = agent_names or set()
     matches = list(_MENTION_RE.finditer(text))
     if not matches:
-        return text, 0
+        return text, 0, []
 
     appendix_parts = []
+    images: list[Image] = []
     seen = set()
     for m in matches:
         rel_path = m.group(1)
@@ -39,9 +45,21 @@ def _resolve_mentions(text: str, cwd: str, agent_names: set[str] | None = None)
         if rel_path in seen:
             continue
         seen.add(rel_path)
-        abs_path = os.path.join(cwd, rel_path)
+        abs_path = rel_path if os.path.isabs(rel_path) else os.path.join(cwd, rel_path)
         if not os.path.isfile(abs_path):
             continue
+
+        ext = os.path.splitext(rel_path)[1].lower()
+        if ext in _IMAGE_EXTENSIONS:
+            try:
+                size = os.path.getsize(abs_path)
+                if size > _IMAGE_MAX_SIZE:
+                    continue
+                images.append(Image(filepath=abs_path, id=rel_path))
+            except OSError:
+                pass
+            continue
+
         try:
             size = os.path.getsize(abs_path)
             with open(abs_path, "r", encoding="utf-8", errors="replace") as f:
@@ -57,9 +75,10 @@ def _resolve_mentions(text: str, cwd: str, agent_names: set[str] | None = None)
         except OSError:
             continue
 
+    attached = len(appendix_parts) + len(images)
     if appendix_parts:
-        return text + "".join(appendix_parts), len(appendix_parts)
-    return text, 0
+        return text + "".join(appendix_parts), attached, images
+    return text, attached, images
 
 
 def _extract_agent_mention(
@@ -206,7 +225,9 @@ class FileMentionCompleter(Completer):
 
             is_dir = os.path.isdir(full_path)
             display_text = rel_prefix + entry + ("/" if is_dir else "")
-            meta = "dir" if is_dir else ""
+            file_ext = os.path.splitext(entry)[1].lower()
+            is_image = not is_dir and file_ext in _IMAGE_EXTENSIONS
+            meta = "dir" if is_dir else ("image" if is_image else "")
 
             yield Completion(
                 display_text,

{aru_code-0.10.0 → aru_code-0.11.0}/aru/runner.py

@@ -43,7 +43,8 @@ class AgentRunResult:
         return f"{self.content}\n\n[Tools]\n{tools_section}"
 
 
-async def run_agent_capture(agent, message: str, session=None, lightweight: bool = False) -> AgentRunResult:
+async def run_agent_capture(agent, message: str, session=None, lightweight: bool = False,
+                           images: list | None = None) -> AgentRunResult:
     """Run agent with async streaming display and parallel tool execution.
 
     Args:
@@ -51,6 +52,7 @@ async def run_agent_capture(agent, message: str, session=None, lightweight: bool
         message: The user message/prompt.
         session: Optional session for history and context.
         lightweight: If True, skip tree/git/plan context and history (for executor steps).
+        images: Optional list of agno.media.Image objects to attach.
 
     Returns:
         AgentRunResult with text output and list of tool call labels.
@@ -116,7 +118,7 @@ async def run_agent_capture(agent, message: str, session=None, lightweight: bool
 
         # Combine: history messages + current enriched message
         if history_messages:
-            history_messages.append(Message(role="user", content=run_message))
+            history_messages.append(Message(role="user", content=run_message, images=images or None))
             agent_input = history_messages
         else:
             agent_input = run_message
@@ -130,7 +132,10 @@ async def run_agent_capture(agent, message: str, session=None, lightweight: bool
             _stall_counter = 0
             _stalled = False
             _STALL_LIMIT = 20
-            async for event in agent.arun(agent_input, stream=True, stream_events=True, yield_run_output=True):
+            arun_kwargs = dict(stream=True, stream_events=True, yield_run_output=True)
+            if isinstance(agent_input, str) and images:
+                arun_kwargs["images"] = images
+            async for event in agent.arun(agent_input, **arun_kwargs):
                 if isinstance(event, RunOutput):
                     run_output = event
                     break

{aru_code-0.10.0 → aru_code-0.11.0}/aru/tools/codebase.py

@@ -1023,8 +1023,19 @@ async def delegate_task(task: str, context: str = "", agent: str = "") -> str:
 
         agent_perm = None
         custom_agent_defs = get_ctx().custom_agent_defs
-        if agent and agent in custom_agent_defs:
-            agent_def = custom_agent_defs[agent]
+        # Agno may pass the caller Agent object instead of a string — coerce to str
+        agent_name = str(agent) if agent and isinstance(agent, str) else ""
+
+        # Print delegation info so the user sees what's happening
+        from rich.console import Console
+        _console = Console()
+        if agent_name and agent_name in custom_agent_defs:
+            _console.print(f"[dim]  → Delegating to agent [bold]{agent_name}[/bold] (task: {task[:80]}{'...' if len(task) > 80 else ''})[/dim]")
+        else:
+            _console.print(f"[dim]  → Delegating to sub-agent #{agent_id} (task: {task[:80]}{'...' if len(task) > 80 else ''})[/dim]")
+
+        if agent_name and agent_name in custom_agent_defs:
+            agent_def = custom_agent_defs[agent_name]
             agent_perm = agent_def.permission
             tools = resolve_tools(agent_def.tools) if agent_def.tools else list(_SUBAGENT_TOOLS)
             tools = [t for t in tools if t is not delegate_task]
@@ -1186,13 +1197,15 @@ def _update_delegate_task_docstring():
     Args:
         task: What the sub-agent should do.
         context: Optional extra context (file paths, constraints).
-        agent: Optional custom agent name to use instead of the generic sub-agent."""
+        agent: Name of a specialized agent to use. ALWAYS prefer a specialized agent when one matches the task."""
 
     custom_agent_defs = get_ctx().custom_agent_defs
     if custom_agent_defs:
-        lines = [f"\n\n    Available specialized agents (use the agent parameter to invoke):"]
+        lines = [f"\n\n    IMPORTANT: When a specialized agent matches the task, you MUST pass its name in the agent parameter."]
+        lines.append(f"    Available specialized agents:")
         for name, agent_def in custom_agent_defs.items():
             lines.append(f"    - agent=\"{name}\": {agent_def.description}")
+        lines.append(f"\n    If no specialized agent fits, omit the agent parameter to use a generic sub-agent.")
         base_doc += "\n".join(lines)
 
     delegate_task.__doc__ = base_doc

{aru_code-0.10.0 → aru_code-0.11.0/aru_code.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aru-code
-Version: 0.10.0
+Version: 0.11.0
 Summary: A Claude Code clone built with Agno agents
 Author-email: Estevao <estevaofon@gmail.com>
 License-Expression: MIT
@@ -55,6 +55,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 
 - **Multi-Agent Architecture** — Specialized agents for planning, execution, and conversation
 - **Interactive CLI** — Streaming responses, multi-line paste, session management
+- **Image Support** — Attach images via `@` mentions for multimodal analysis (Claude, GPT-4o, Gemini)
 - **16 Integrated Tools** — File operations, code search, shell, web search, task delegation
 - **Task Planning** — Break down complex tasks into steps with automatic execution
 - **Multi-Provider** — Anthropic, OpenAI, Ollama, Groq, OpenRouter, DeepSeek, and others via custom configuration
@@ -69,7 +70,7 @@ An intelligent coding assistant for the terminal, powered by LLMs and [Agno](htt
 pip install aru-code
 ```
 
-> **Requirements:** Python 3.13+
+> **Requirements:** Python 3.11+
 
 ### 2. Configure the API Key
 
@@ -131,6 +132,23 @@ aru> ! pytest tests/ -v
 aru> /model ollama/codellama
 ```
 
+### Image Support
+
+Attach images to your messages using the same `@` mention syntax used for files. Aru detects image files by extension and sends them to the LLM as visual content for multimodal analysis.
+
+**Supported formats:** `.png`, `.jpg`, `.jpeg`, `.gif`, `.webp`, `.bmp`
+
+```
+aru> describe @screenshot.png
+aru> compare @before.png and @after.png
+aru> review @code.py and explain the diagram in @architecture.png
+aru> analyze @D:/full/path/to/image.jpg
+```
+
+Images are sent natively to the model via the provider's multimodal API — no base64 text is injected into the conversation. Works with any multimodal model (Claude Opus/Sonnet, GPT-4o, Gemini, etc.). The autocomplete shows an `[image]` label for image files.
+
+> **Note:** Images require a multimodal model. Local models via Ollama may not support image input. Maximum file size: 20MB.
+
 ## Configuration
 
 ### Models and Providers
@@ -374,11 +392,22 @@ performance, and readability. Do NOT modify files.
 
 #### Invocation
 
+There are three ways to invoke a custom agent:
+
+| Method | Syntax | When to use |
+|--------|--------|-------------|
+| **Slash command** | `/reviewer src/auth.py` | Directly invoke a `primary` agent by name |
+| **@mention** | `@reviewer check this function` | Mention an agent anywhere in your message |
+| **delegate_task** | Automatic (subagents only) | Subagent names and descriptions are injected into the `delegate_task` tool description, so the LLM sees them and can call `delegate_task(task="...", agent="name")` on its own when it judges the task fits |
+
 ```
-aru> /reviewer src/auth.py        # invoke by slash + filename (without .md)
-aru> /agents                       # list all custom agents
+aru> /reviewer src/auth.py           # slash command (primary agents)
+aru> @reviewer check the auth module  # @mention (primary or subagent)
+aru> /agents                          # list all custom agents
 ```
 
+> **Note:** Slash commands (`/name`) are only available for `primary` agents — subagents are blocked with a warning. `@mention` works for any agent regardless of mode. Subagents can be invoked in two ways: automatically by the LLM via `delegate_task`, or manually by the user via `@name`.
+
 #### Discovery paths
 
 Agents are discovered from multiple locations (later overrides earlier):

{aru_code-0.10.0 → aru_code-0.11.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "aru-code"
-version = "0.10.0"
+version = "0.11.0"
 description = "A Claude Code clone built with Agno agents"
 readme = "README.md"
 license = "MIT"

{aru_code-0.10.0 → aru_code-0.11.0}/tests/test_cli.py

@@ -46,32 +46,32 @@ class TestSanitizeInput:
 
 class TestResolveMentions:
     def test_no_mentions(self, tmp_path):
-        result, count = _resolve_mentions("hello world", str(tmp_path))
+        result, count, _imgs = _resolve_mentions("hello world", str(tmp_path))
         assert result == "hello world"
         assert count == 0
 
     def test_resolves_file_mention(self, tmp_path):
         (tmp_path / "config.py").write_text("DEBUG = True")
-        result, count = _resolve_mentions("check @config.py", str(tmp_path))
+        result, count, _imgs = _resolve_mentions("check @config.py", str(tmp_path))
         assert "DEBUG = True" in result
         assert "Contents of config.py" in result
         assert count == 1
 
     def test_nonexistent_file_ignored(self, tmp_path):
-        result, count = _resolve_mentions("check @missing.py", str(tmp_path))
+        result, count, _imgs = _resolve_mentions("check @missing.py", str(tmp_path))
         assert result == "check @missing.py"
         assert count == 0
 
     def test_deduplicates_mentions(self, tmp_path):
         (tmp_path / "file.py").write_text("code")
-        result, count = _resolve_mentions("@file.py and @file.py", str(tmp_path))
+        result, count, _imgs = _resolve_mentions("@file.py and @file.py", str(tmp_path))
         assert result.count("Contents of file.py") == 1
         assert count == 1
 
     def test_multiple_files(self, tmp_path):
         (tmp_path / "a.py").write_text("aaa")
         (tmp_path / "b.py").write_text("bbb")
-        result, count = _resolve_mentions("@a.py and @b.py", str(tmp_path))
+        result, count, _imgs = _resolve_mentions("@a.py and @b.py", str(tmp_path))
         assert "Contents of a.py" in result
         assert "Contents of b.py" in result
         assert count == 2

{aru_code-0.10.0 → aru_code-0.11.0}/tests/test_cli_completers.py

@@ -7,13 +7,17 @@ from unittest.mock import Mock, patch
 import pytest
 from prompt_toolkit.document import Document
 
+from agno.media import Image
+
 from aru.cli import (
     SlashCommandCompleter,
     FileMentionCompleter,
     AruCompleter,
     SLASH_COMMANDS,
     _extract_agent_mention,
+    _resolve_mentions,
 )
+from aru.completers import _IMAGE_EXTENSIONS
 from aru.config import CustomAgent, CustomCommand
 
 
@@ -593,4 +597,96 @@ class TestExtractAgentMention:
     def test_agent_not_matched_if_attached_to_word(self):
         agents = self._make_agents()
         # @reviewer preceded by non-whitespace should not match
-        assert _extract_agent_mention("email@reviewer", agents) is None
+        assert _extract_agent_mention("email@reviewer", agents) is None
+
+
+# ── Image mention support ──────────────────────────────────────────
+
+
+class TestImageMentions:
+    """Tests for image file detection in @mentions."""
+
+    def test_resolve_mentions_returns_three_tuple(self, tmp_path):
+        result = _resolve_mentions("hello", str(tmp_path))
+        assert len(result) == 3
+        text, count, images = result
+        assert text == "hello"
+        assert count == 0
+        assert images == []
+
+    def test_resolve_mentions_image_file(self, tmp_path):
+        img = tmp_path / "screenshot.png"
+        img.write_bytes(b"\x89PNG\r\n\x1a\n" + b"\x00" * 100)
+
+        text, count, images = _resolve_mentions(
+            "analyze @screenshot.png", str(tmp_path)
+        )
+        assert count == 1
+        assert len(images) == 1
+        assert isinstance(images[0], Image)
+        assert images[0].id == "screenshot.png"
+        # Image content should NOT be appended as text
+        assert "```" not in text
+
+    def test_resolve_mentions_mixed_files_and_images(self, tmp_path):
+        (tmp_path / "code.py").write_text("print('hello')", encoding="utf-8")
+        (tmp_path / "diagram.jpg").write_bytes(b"\xff\xd8\xff" + b"\x00" * 100)
+
+        text, count, images = _resolve_mentions(
+            "review @code.py and @diagram.jpg", str(tmp_path)
+        )
+        assert count == 2
+        assert len(images) == 1
+        assert images[0].id == "diagram.jpg"
+        # Text file content should be appended
+        assert "print('hello')" in text
+
+    def test_resolve_mentions_multiple_images(self, tmp_path):
+        (tmp_path / "a.png").write_bytes(b"\x89PNG" + b"\x00" * 100)
+        (tmp_path / "b.webp").write_bytes(b"RIFF" + b"\x00" * 100)
+
+        text, count, images = _resolve_mentions(
+            "compare @a.png @b.webp", str(tmp_path)
+        )
+        assert count == 2
+        assert len(images) == 2
+
+    def test_resolve_mentions_image_too_large(self, tmp_path):
+        img = tmp_path / "huge.png"
+        # Write just over the 20MB limit header
+        img.write_bytes(b"\x89PNG" + b"\x00" * (20 * 1024 * 1024 + 1))
+
+        text, count, images = _resolve_mentions(
+            "analyze @huge.png", str(tmp_path)
+        )
+        assert count == 0
+        assert len(images) == 0
+
+    def test_resolve_mentions_all_image_extensions(self, tmp_path):
+        for ext in _IMAGE_EXTENSIONS:
+            fname = f"test{ext}"
+            (tmp_path / fname).write_bytes(b"\x00" * 100)
+
+        mentions = " ".join(f"@test{ext}" for ext in _IMAGE_EXTENSIONS)
+        text, count, images = _resolve_mentions(mentions, str(tmp_path))
+        assert len(images) == len(_IMAGE_EXTENSIONS)
+
+    def test_image_completer_shows_image_metadata(self, tmp_path):
+        (tmp_path / "photo.png").touch()
+        (tmp_path / "code.py").touch()
+
+        completer = FileMentionCompleter()
+        doc = Document("@")
+        with patch("os.getcwd", return_value=str(tmp_path)):
+            with patch("aru.tools.gitignore.is_ignored", return_value=False):
+                completions = list(completer.get_completions(doc, Mock()))
+
+        by_name = {c.text: c for c in completions}
+        assert "photo.png" in by_name
+        assert "code.py" in by_name
+        # Image should have "image" in metadata
+        photo_meta = str(by_name["photo.png"].display_meta)
+        assert "image" in photo_meta
+        # Code file should NOT have "image" in metadata
+        code_meta = str(by_name["code.py"].display_meta)
+        assert "image" not in code_meta

aru_code-0.10.0/aru/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.10.0"