tunacode-cli 0.0.33__tar.gz → 0.0.35__tar.gz

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/CLAUDE.md

@@ -10,251 +10,18 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 # Install development environment (recommended approach)
 ./scripts/setup_dev_env.sh    # Creates fresh venv, installs deps, verifies setup
 
-# You must always follow the flow for working
- Project structure:
-  agent-tools/
-  ├── wakeup.sh          # Read memory bank
-  ├── scratchpad.sh      # Task logging
-  ├── check_workflow.sh  # Simple verification
-  ├── bankctl.sh         # Memory bank control
-  └── WORKFLOW_GUIDE.md  # Complete workflow documentation
-
-  memory-bank/
-  ├── project_brief.md
-  ├── tech_context.md
-  ├── product_context.md
-  ├── current_state_summary.md
-  └── progress_overview.md
-
-🚀 Quick start:
-  1. Edit memory-bank/*.md files with your project details
-  2. ./wakeup.sh                    # Read current context
-  3. ./scratchpad.sh start 'Task'   # Begin new task
-  4. ./scratchpad.sh step 'Action'  # Log progress
-  5. ./scratchpad.sh close 'Done'   # Complete task
-
-📖 Full guide: cat agent-tools/WORKFLOW_GUIDE.md
-<?xml version="1.0" encoding="UTF-8"?>
-<system_prompt>
-
-###Instruction###
-
-You are an expert software engineering assistant equipped with specialized bash tools for memory management and task tracking. Your primary goal is to maintain persistent context across sessions while following a structured workflow.
-
-You MUST use these tools proactively and frequently. You will be penalized for failing to use appropriate tools when they would improve task outcomes.
-
-<role>Expert Software Engineering Assistant with Memory Management Tools</role>
-
-<available_tools>
-1. wakeup.sh - Read memory bank to regain project context
-2. scratchpad.sh - Task logging and progress tracking
-3. check_workflow.sh - Simple verification check
-4. bankctl.sh - Memory bank initialization and management
-</available_tools>
-
-<critical_requirements>
-- Think step by step when approaching any task
-- Always run wakeup.sh at the start of a new session to regain context
-- Use scratchpad.sh for EVERY task to maintain detailed work logs
-- Update memory-bank/current_state_summary.md after completing tasks
-- Occasionally run check_workflow.sh to verify nothing was missed
-- Ensure that your approach is unbiased and does not rely on stereotypes
-</critical_requirements>
-
-###Example###
-
-<example_workflow>
-User: "Help me implement a new user registration feature"
-
-CORRECT APPROACH:
-1. ./wakeup.sh (read memory bank to understand project context)
-2. ./scratchpad.sh start "Implement user registration feature"
-3. ./scratchpad.sh plan "1. Create user model 2. Design API endpoint 3. Add validation 4. Write tests"
-4. ./scratchpad.sh step "Created User model in models.py with email, username, password_hash"
-5. ./scratchpad.sh step "Implemented POST /register endpoint with input validation"
-6. ./scratchpad.sh step "Added password hashing using bcrypt"
-7. ./scratchpad.sh step "Wrote unit tests for registration flow"
-8. ./scratchpad.sh close "User registration feature complete"
-9. Update memory-bank/current_state_summary.md with session outcome
-10. ./check_workflow.sh (verify workflow was followed)
-
-INCORRECT APPROACH:
-- Starting work without reading memory bank
-- Making changes without tracking steps in scratchpad
-- Not updating current_state_summary.md after task completion
-- Never checking if workflow was properly followed
-</example_workflow>
-
-###Guidelines###
-
-<wakeup_usage>
-WHEN TO USE:
-- At the start of EVERY new session
-- When returning to a project after any break
-- To understand project context and current state
-
-OUTPUT:
-- Reads all memory bank files in priority order
-- Shows current_state_summary.md first (most important)
-- Displays project brief, technical context, product context, and progress
-
-You MUST:
-- Always run wakeup.sh before starting any work
-- Pay special attention to current_state_summary.md
-- Use the context to inform your approach
-</wakeup_usage>
-
-<scratchpad_usage>
-WHEN TO USE:
-- For EVERY task, regardless of complexity
-- Even for single-step tasks (maintains history)
-- When exploring, debugging, or implementing features
-
-COMMANDS:
-- start "task_name": Begin new task tracking
-- plan "plan_details": Document your approach
-- step "action_taken": Log each action/decision
-- close "completion_message": Archive the task
-
-You MUST:
-- Start scratchpad for every task
-- Log detailed steps as you work
-- Close and archive when complete
-- Note: close command auto-sanitizes filenames
-</scratchpad_usage>
-
-<check_workflow_usage>
-WHEN TO USE:
-- After completing a few tasks
-- When you want to verify workflow compliance
-- Periodically to ensure nothing was missed
-
-OUTPUT:
-- Shows when memory bank was last updated
-- Lists recent archived scratchpads
-- Displays current state summary
-
-You SHOULD:
-- Run this occasionally (not after every single task)
-- Use it as a sanity check for workflow adherence
-- Pay attention if updates are getting stale
-</check_workflow_usage>
-
-<bankctl_usage>
-WHEN TO USE:
-- First time setup of a project
-- When memory bank structure needs initialization
-- For memory bank maintenance tasks
-
-COMMANDS:
-- init: Initialize memory bank structure
-- Other commands vary by implementation
-
-You MUST:
-- Use bankctl.sh init for new projects
-- Ensure memory bank exists before using other tools
-</bankctl_usage>
-
-<memory_bank_structure>
-CORE FILES:
-1. project_brief.md - What & why of the project
-2. tech_context.md - Technical decisions & architecture
-3. product_context.md - User experience goals
-4. current_state_summary.md - CRITICAL: Latest state & next steps
-5. progress_overview.md - Feature/task tracker
-
-UPDATE STRATEGY:
-- current_state_summary.md: Update after EVERY session
-- progress_overview.md: Update when features complete
-- Other files: Update only when fundamentals change
-
-You MUST:
-- Keep current_state_summary.md concise but complete
-- Include session outcomes and immediate next steps
-- Archive detailed logs in scratchpad, not memory bank
-</memory_bank_structure>
-
-###Workflow_Patterns###
-
-<pattern name="new_session_startup">
-1. ./wakeup.sh
-2. Review current_state_summary.md carefully
-3. Identify immediate next objectives
-4. ./scratchpad.sh start "[next_task_from_summary]"
-5. Continue with task implementation
-</pattern>
-
-<pattern name="feature_implementation">
-1. ./wakeup.sh
-2. ./scratchpad.sh start "Implement [feature_name]"
-3. ./scratchpad.sh plan "Steps: 1. [step1] 2. [step2] 3. [step3]"
-4. ./scratchpad.sh step "Completed [specific action]"
-5. [continue logging each step]
-6. ./scratchpad.sh close "[feature_name] implementation complete"
-7. Update memory-bank/current_state_summary.md
-8. Update memory-bank/progress_overview.md
-9. ./check_workflow.sh (occasionally, to verify)
-</pattern>
-
-<pattern name="debugging_session">
-1. ./wakeup.sh
-2. ./scratchpad.sh start "Debug [issue_description]"
-3. ./scratchpad.sh step "Reproduced issue: [details]"
-4. ./scratchpad.sh step "Identified root cause: [cause]"
-5. ./scratchpad.sh step "Applied fix: [solution]"
-6. ./scratchpad.sh step "Verified fix works"
-7. ./scratchpad.sh close "Fixed [issue_description]"
-8. Update memory-bank/current_state_summary.md
-</pattern>
-
-<pattern name="project_initialization">
-1. ./bankctl.sh init
-2. Edit memory-bank/project_brief.md
-3. Edit memory-bank/tech_context.md
-4. Edit memory-bank/product_context.md
-5. Edit memory-bank/current_state_summary.md
-6. Edit memory-bank/progress_overview.md
-7. ./wakeup.sh (verify setup)
-</pattern>
-
-###Penalties###
-
-You will be penalized for:
-- Not running wakeup.sh at session start
-- Starting any task without scratchpad.sh
-- Failing to update current_state_summary.md after tasks
-- Not archiving completed scratchpads
-- Keeping detailed logs in memory bank instead of scratchpad
-- Never running check_workflow.sh to verify compliance
-
-###Output_Format###
-
-When using tools, always show:
-1. The exact command being executed
-2. Brief explanation of why you're using it
-3. Key findings or results
-
-###Memory_Management_Philosophy###
-
-This workflow is designed for agents that experience complete memory loss between sessions. The system provides:
-
-1. **Memory Bank** - Persistent, summarized knowledge base
-   - Project context and goals
-   - Current state and next steps
-   - High-level progress tracking
-
-2. **Scratchpad** - Detailed, temporary work logs
-   - Step-by-step task documentation
-   - Decisions and observations
-   - Archived after completion
-
-The key is maintaining clear separation between long-term strategic memory (Memory Bank) and short-term operational memory (Scratchpad).
-
-Answer questions in a natural, human-like manner while maintaining technical accuracy.
-
-I'm going to tip $200000 for exceptional workflow adherence that demonstrates mastery of memory management!
-
-</system_prompt>
+# You must always follow the agent tooling workflow specified in agent_tools_prompt.xml
+
+When the directory "llm-agent-tools" exists, you MUST use this tooling flow and will be penalized if you don't:
+
+🚀 Available Tools:
+1. scratchpad-multi.sh - Task tracking with multi-agent support (use --agent <name> for agent-specific work)
+2. knowledge.sh - Knowledge base with private/shared storage (use --agent <name> for agent-specific work) 
+3. codemap.sh - Lightweight code intelligence for instant roadmaps
+4. context.sh - Context gathering for debugging issues
+5. researcher.sh - Online research and multimodal analysis via OpenRouter API
+
+📖 Full workflow guide: see llm-agent-tools/agent_tools_prompt.xml
 
 For new feature YOU MUST folow this flow
 

{tunacode_cli-0.0.33/src/tunacode_cli.egg-info → tunacode_cli-0.0.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tunacode-cli
-Version: 0.0.33
+Version: 0.0.35
 Summary: Your agentic CLI developer.
 Author-email: larock22 <noreply@github.com>
 License-Expression: MIT
@@ -40,6 +40,7 @@ Dynamic: license-file
 <div align="center">
 
 [![PyPI version](https://badge.fury.io/py/tunacode-cli.svg)](https://badge.fury.io/py/tunacode-cli)
+[![Downloads](https://pepy.tech/badge/tunacode-cli)](https://pepy.tech/project/tunacode-cli)
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
@@ -78,6 +79,17 @@ tunacode --model "openrouter:openai/gpt-4o" --key "sk-or-your-openrouter-key"
 
 Your config is saved to `~/.config/tunacode.json` (edit directly with `nvim ~/.config/tunacode.json`)
 
+### Recommended Models
+
+Based on extensive testing, these models provide the best performance:
+- `google/gemini-2.5-pro` - Excellent for complex reasoning
+- `openai/gpt-4.1` - Strong general-purpose model
+- `deepseek/deepseek-r1-0528` - Great for code generation
+- `openai/gpt-4.1-mini` - Fast and cost-effective
+- `anthropic/claude-4-sonnet-20250522` - Superior context handling
+
+*Note: Formal evaluations coming soon. Any model can work, but these have shown the best results in practice.*
+
 ## Start Coding
 
 ```bash
@@ -105,6 +117,13 @@ TunaCode leverages parallel execution for read-only operations, achieving **3x f
 
 Multiple file reads, directory listings, and searches execute concurrently using async I/O, making code exploration significantly faster.
 
+## Features in Development
+
+- **Streaming UI**: Currently working on implementing streaming responses for better user experience
+- **Bug Fixes**: Actively addressing issues - please report any bugs you encounter!
+
+*Note: While the tool is fully functional, we're focusing on stability and core features before optimizing for speed.*
+
 ## Safety First
 
 ⚠️ **Important**: TunaCode can modify your codebase. Always:

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/README.md

@@ -3,6 +3,7 @@
 <div align="center">
 
 [![PyPI version](https://badge.fury.io/py/tunacode-cli.svg)](https://badge.fury.io/py/tunacode-cli)
+[![Downloads](https://pepy.tech/badge/tunacode-cli)](https://pepy.tech/project/tunacode-cli)
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
@@ -41,6 +42,17 @@ tunacode --model "openrouter:openai/gpt-4o" --key "sk-or-your-openrouter-key"
 
 Your config is saved to `~/.config/tunacode.json` (edit directly with `nvim ~/.config/tunacode.json`)
 
+### Recommended Models
+
+Based on extensive testing, these models provide the best performance:
+- `google/gemini-2.5-pro` - Excellent for complex reasoning
+- `openai/gpt-4.1` - Strong general-purpose model
+- `deepseek/deepseek-r1-0528` - Great for code generation
+- `openai/gpt-4.1-mini` - Fast and cost-effective
+- `anthropic/claude-4-sonnet-20250522` - Superior context handling
+
+*Note: Formal evaluations coming soon. Any model can work, but these have shown the best results in practice.*
+
 ## Start Coding
 
 ```bash
@@ -68,6 +80,13 @@ TunaCode leverages parallel execution for read-only operations, achieving **3x f
 
 Multiple file reads, directory listings, and searches execute concurrently using async I/O, making code exploration significantly faster.
 
+## Features in Development
+
+- **Streaming UI**: Currently working on implementing streaming responses for better user experience
+- **Bug Fixes**: Actively addressing issues - please report any bugs you encounter!
+
+*Note: While the tool is fully functional, we're focusing on stability and core features before optimizing for speed.*
+
 ## Safety First
 
 ⚠️ **Important**: TunaCode can modify your codebase. Always:

tunacode_cli-0.0.35/TUNACODE.md

@@ -0,0 +1,27 @@
+# TUNACODE.md
+
+## Build Commands
+- Run all tests: `make test`
+- Run single test: `pytest tests/test_file.py::test_name`
+- Run quick tests: `pytest -m "not slow"`
+- Lint code: `make lint`
+- Build package: `make build`
+- Clean artifacts: `make clean`
+
+## Code Style
+- Use type hints for all function signatures
+- Prefer guard clauses over nested conditionals
+- Keep functions focused and under 50 lines
+- Use descriptive variable names
+- Follow PEP 8 conventions
+- Maximum line length: 120 characters
+- Use black for formatting
+- Use isort for imports
+- Use snake_case for variables and functions
+- Use PascalCase for classes
+- Prefer explicit error handling, avoid bare excepts
+
+## Architecture Notes
+- Agent creation loads this file and appends to system prompt
+- Context is loaded synchronously to avoid event loop issues
+- TUNACODE.md is walked up directory tree (closest first)

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "tunacode-cli"
-version = "0.0.33"
+version = "0.0.35"
 description = "Your agentic CLI developer."
 keywords = ["cli", "agent", "development", "automation"]
 readme = "README.md"

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/src/tunacode/cli/commands.py

@@ -6,7 +6,7 @@ from enum import Enum
 from typing import Any, Dict, List, Optional, Type
 
 from .. import utils
-from ..exceptions import ValidationError
+from ..exceptions import ConfigurationError, ValidationError
 from ..types import CommandArgs, CommandContext, CommandResult, ProcessRequestCallback
 from ..ui import console as ui
 
@@ -181,26 +181,29 @@ class IterationsCommand(SimpleCommand):
 
     async def execute(self, args: List[str], context: CommandContext) -> None:
         state = context.state_manager.session
-        if args:
-            try:
-                new_limit = int(args[0])
-                if new_limit < 1 or new_limit > 100:
-                    await ui.error("Iterations must be between 1 and 100")
-                    return
-
-                # Update the user config
-                if "settings" not in state.user_config:
-                    state.user_config["settings"] = {}
-                state.user_config["settings"]["max_iterations"] = new_limit
-
-                await ui.success(f"Maximum iterations set to {new_limit}")
-                await ui.muted("Higher values allow more complex reasoning but may be slower")
-            except ValueError:
-                await ui.error("Please provide a valid number")
-        else:
+
+        # Guard clause - handle "no args" case first and return early
+        if not args:
             current = state.user_config.get("settings", {}).get("max_iterations", 40)
             await ui.info(f"Current maximum iterations: {current}")
             await ui.muted("Usage: /iterations <number> (1-100)")
+            return
+
+        # update the logic to not be as nested messely, the above guars needing to get as messy
+        try:
+            new_limit = int(args[0])
+            if new_limit < 1 or new_limit > 100:
+                await ui.error("Iterations must be between 1 and 100")
+                return
+
+            # Update the user config
+            if "settings" not in state.user_config:
+                state.user_config["settings"] = {}
+            state.user_config["settings"]["max_iterations"] = new_limit
+
+            await ui.success(f"Maximum iterations set to {new_limit}")
+        except ValueError:
+            await ui.error("Please provide a valid number")
 
 
 class ClearCommand(SimpleCommand):
@@ -288,7 +291,9 @@ class ParseToolsCommand(SimpleCommand):
 
                         try:
                             await extract_and_execute_tool_calls(
-                                part.content, tool_callback_with_state, context.state_manager
+                                part.content,
+                                tool_callback_with_state,
+                                context.state_manager,
                             )
                             await ui.success("JSON tool parsing completed")
                             found_content = True
@@ -524,7 +529,8 @@ class UpdateCommand(SimpleCommand):
                 result = subprocess.run(
                     ["pipx", "list"], capture_output=True, text=True, timeout=10
                 )
-                if "tunacode" in result.stdout.lower():
+                pipx_installed = "tunacode" in result.stdout.lower()
+                if pipx_installed:
                     installation_method = "pipx"
             except (subprocess.TimeoutExpired, subprocess.CalledProcessError):
                 pass
@@ -555,12 +561,22 @@ class UpdateCommand(SimpleCommand):
             if installation_method == "pipx":
                 await ui.info("Updating via pipx...")
                 result = subprocess.run(
-                    ["pipx", "upgrade", "tunacode"], capture_output=True, text=True, timeout=60
+                    ["pipx", "upgrade", "tunacode"],
+                    capture_output=True,
+                    text=True,
+                    timeout=60,
                 )
             else:  # pip
                 await ui.info("Updating via pip...")
                 result = subprocess.run(
-                    [sys.executable, "-m", "pip", "install", "--upgrade", "tunacode-cli"],
+                    [
+                        sys.executable,
+                        "-m",
+                        "pip",
+                        "install",
+                        "--upgrade",
+                        "tunacode-cli",
+                    ],
                     capture_output=True,
                     text=True,
                     timeout=60,
@@ -628,9 +644,13 @@ class ModelCommand(SimpleCommand):
 
         # Check if setting as default
         if len(args) > 1 and args[1] == "default":
-            utils.user_configuration.set_default_model(model_name, context.state_manager)
-            await ui.muted("Updating default model")
-            return "restart"
+            try:
+                utils.user_configuration.set_default_model(model_name, context.state_manager)
+                await ui.muted("Updating default model")
+                return "restart"
+            except ConfigurationError as e:
+                await ui.error(str(e))
+                return None
 
         # Show success message with the new model
         await ui.success(f"Switched to model: {model_name}")
@@ -669,6 +689,35 @@ class CommandFactory:
                 setattr(self.dependencies, key, value)
 
 
+class InitCommand(SimpleCommand):
+    """Creates or updates TUNACODE.md with project-specific context."""
+
+    spec = CommandSpec(
+        name="/init",
+        aliases=[],
+        description="Analyze codebase and create/update TUNACODE.md file",
+        category=CommandCategory.DEVELOPMENT,
+    )
+
+    async def execute(self, args, context: CommandContext) -> CommandResult:
+        """Execute the init command."""
+        # Minimal implementation to make test pass
+        prompt = """Please analyze this codebase and create a TUNACODE.md file containing:
+1. Build/lint/test commands - especially for running a single test
+2. Code style guidelines including imports, formatting, types, naming conventions, error handling, etc.
+
+The file you create will be given to agentic coding agents (such as yourself) that operate in this repository.
+Make it about 20 lines long.
+If there's already a TUNACODE.md, improve it.
+If there are Cursor rules (in .cursor/rules/ or .cursorrules) or Copilot rules (in .github/copilot-instructions.md),
+make sure to include them."""
+
+        # Call the agent to analyze and create/update the file
+        await context.process_request(prompt, context.state_manager)
+
+        return None
+
+
 class CommandRegistry:
     """Registry for managing commands with auto-discovery and categories."""
 
@@ -726,6 +775,7 @@ class CommandRegistry:
             BranchCommand,
             CompactCommand,
             ModelCommand,
+            InitCommand,
         ]
 
         # Register all discovered commands

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/src/tunacode/cli/repl.py

@@ -22,6 +22,7 @@ from tunacode.core.tool_handler import ToolHandler
 from tunacode.exceptions import AgentError, UserAbortError, ValidationError
 from tunacode.ui import console as ui
 from tunacode.ui.tool_ui import ToolUI
+from tunacode.utils.security import CommandSecurityError, safe_subprocess_run
 
 from ..types import CommandContext, CommandResult, StateManager, ToolArgs
 from .commands import CommandRegistry
@@ -320,13 +321,24 @@ async def repl(state_manager: StateManager):
                 def run_shell():
                     try:
                         if command:
-                            result = subprocess.run(command, shell=True, capture_output=False)
-                            if result.returncode != 0:
-                                # Use print directly since we're in a terminal context
-                                print(f"\nCommand exited with code {result.returncode}")
+                            # Use secure subprocess execution for shell commands
+                            # Note: User shell commands are inherently risky but this is by design
+                            # We validate but allow shell features since it's explicit user intent
+                            try:
+                                result = safe_subprocess_run(
+                                    command,
+                                    shell=True,
+                                    validate=True,  # Still validate for basic safety
+                                    capture_output=False,
+                                )
+                                if result.returncode != 0:
+                                    print(f"\nCommand exited with code {result.returncode}")
+                            except CommandSecurityError as e:
+                                print(f"\nSecurity validation failed: {str(e)}")
+                                print("If you need to run this command, please ensure it's safe.")
                         else:
                             shell = os.environ.get("SHELL", "bash")
-                            subprocess.run(shell)
+                            subprocess.run(shell)  # Interactive shell is safe
                     except Exception as e:
                         print(f"\nShell command failed: {str(e)}")
 

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/src/tunacode/constants.py

@@ -7,7 +7,7 @@ Centralizes all magic strings, UI text, error messages, and application constant
 
 # Application info
 APP_NAME = "TunaCode"
-APP_VERSION = "0.0.33"
+APP_VERSION = "0.0.35"
 
 # File patterns
 GUIDE_FILE_PATTERN = "{name}.md"

{tunacode_cli-0.0.33 → tunacode_cli-0.0.35}/src/tunacode/core/agents/main.py

@@ -457,6 +457,26 @@ def get_or_create_agent(model: ModelName, state_manager: StateManager) -> Pydant
                 # Use a default system prompt if neither file exists
                 system_prompt = "You are a helpful AI assistant for software development tasks."
 
+        # Load TUNACODE.md context
+        # Use sync version of get_code_style to avoid nested event loop issues
+        try:
+            from pathlib import Path as PathlibPath
+
+            tunacode_path = PathlibPath.cwd() / "TUNACODE.md"
+            if tunacode_path.exists():
+                tunacode_content = tunacode_path.read_text(encoding="utf-8")
+                if tunacode_content.strip():
+                    # Log that we found TUNACODE.md
+                    print("📄 TUNACODE.md located: Loading context...")
+
+                    system_prompt += "\n\n# Project Context from TUNACODE.md\n" + tunacode_content
+            else:
+                # Log that TUNACODE.md was not found
+                print("📄 TUNACODE.md not found: Using default context")
+        except Exception:
+            # Ignore errors loading TUNACODE.md
+            pass
+
         state_manager.session.agents[model] = Agent(
             model=model,
             system_prompt=system_prompt,
@@ -670,6 +690,14 @@ async def process_request(
     # Create a request-level buffer for batching read-only tools across nodes
     tool_buffer = ToolBuffer()
 
+    # Show TUNACODE.md preview if it was loaded and thoughts are enabled
+    if state_manager.session.show_thoughts and hasattr(state_manager, "tunacode_preview"):
+        from tunacode.ui import console as ui
+
+        await ui.muted(state_manager.tunacode_preview)
+        # Clear the preview after displaying it once
+        delattr(state_manager, "tunacode_preview")
+
     # Show what we're sending to the API when thoughts are enabled
     if state_manager.session.show_thoughts:
         from tunacode.ui import console as ui