tinyagent-py 0.0.15__py3-none-any.whl → 0.0.16__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -10,7 +10,10 @@ from tinyagent.hooks.rich_code_ui_callback import RichCodeUICallback
10
10
  from tinyagent.hooks.jupyter_notebook_callback import JupyterNotebookCallback
11
11
  from .providers.base import CodeExecutionProvider
12
12
  from .providers.modal_provider import ModalProvider
13
+ from .providers.seatbelt_provider import SeatbeltProvider
13
14
  from .helper import translate_tool_for_code_agent, load_template, render_system_prompt, prompt_code_example, prompt_qwen_helper
15
+ from .utils import truncate_output, format_truncation_message
16
+ import datetime
14
17
 
15
18
 
16
19
  DEFAULT_SUMMARY_SYSTEM_PROMPT = (
@@ -26,7 +29,16 @@ class TinyCodeAgent:
26
29
  A TinyAgent specialized for code execution tasks.
27
30
 
28
31
  This class provides a high-level interface for creating agents that can execute
29
- Python code using various providers (Modal, Docker, local execution, etc.).
32
+ Python code using various providers (Modal, SeatbeltProvider for macOS sandboxing, etc.).
33
+
34
+ Features include:
35
+ - Code execution in sandboxed environments
36
+ - Shell command execution with safety checks
37
+ - Environment variable management (SeatbeltProvider)
38
+ - File system access controls
39
+ - Memory management and conversation summarization
40
+ - Git checkpoint automation
41
+ - Output truncation controls
30
42
  """
31
43
 
32
44
  def __init__(
@@ -48,6 +60,8 @@ class TinyCodeAgent:
48
60
  default_workdir: Optional[str] = None,
49
61
  summary_config: Optional[Dict[str, Any]] = None,
50
62
  ui: Optional[str] = None,
63
+ truncation_config: Optional[Dict[str, Any]] = None,
64
+ auto_git_checkpoint: bool = False,
51
65
  **agent_kwargs
52
66
  ):
53
67
  """
@@ -72,7 +86,33 @@ class TinyCodeAgent:
72
86
  default_workdir: Default working directory for shell commands. If None, the current working directory is used.
73
87
  summary_config: Optional configuration for generating conversation summaries
74
88
  ui: The user interface callback to use ('rich', 'jupyter', or None).
89
+ truncation_config: Configuration for output truncation (max_tokens, max_lines)
90
+ auto_git_checkpoint: If True, automatically create git checkpoints after each successful shell command
75
91
  **agent_kwargs: Additional arguments passed to TinyAgent
92
+
93
+ Provider Config Options:
94
+ For SeatbeltProvider:
95
+ - seatbelt_profile: String containing seatbelt profile rules
96
+ - seatbelt_profile_path: Path to a file containing seatbelt profile rules
97
+ - python_env_path: Path to the Python environment to use
98
+ - bypass_shell_safety: If True, bypass shell command safety checks (default: True for seatbelt)
99
+ - additional_safe_shell_commands: Additional shell commands to consider safe
100
+ - additional_safe_control_operators: Additional shell control operators to consider safe
101
+ - additional_read_dirs: List of additional directories to allow read access to
102
+ - additional_write_dirs: List of additional directories to allow write access to
103
+ - environment_variables: Dictionary of environment variables to make available in the sandbox
104
+
105
+ For ModalProvider:
106
+ - pip_packages: List of additional Python packages to install
107
+ - authorized_imports: List of authorized Python imports
108
+ - bypass_shell_safety: If True, bypass shell command safety checks (default: False for modal)
109
+ - additional_safe_shell_commands: Additional shell commands to consider safe
110
+ - additional_safe_control_operators: Additional shell control operators to consider safe
111
+
112
+ Truncation Config Options:
113
+ - max_tokens: Maximum number of tokens to keep in output (default: 3000)
114
+ - max_lines: Maximum number of lines to keep in output (default: 250)
115
+ - enabled: Whether truncation is enabled (default: True)
76
116
  """
77
117
  self.model = model
78
118
  self.api_key = api_key
@@ -87,6 +127,15 @@ class TinyCodeAgent:
87
127
  self.provider = provider # Store provider type for reuse
88
128
  self.check_string_obfuscation = check_string_obfuscation
89
129
  self.default_workdir = default_workdir or os.getcwd() # Default to current working directory if not specified
130
+ self.auto_git_checkpoint = auto_git_checkpoint # Enable/disable automatic git checkpoints
131
+
132
+ # Set up truncation configuration with defaults
133
+ default_truncation = {
134
+ "max_tokens": 3000,
135
+ "max_lines": 250,
136
+ "enabled": True
137
+ }
138
+ self.truncation_config = {**default_truncation, **(truncation_config or {})}
90
139
 
91
140
  # Create the code execution provider
92
141
  self.code_provider = self._create_provider(provider, self.provider_config)
@@ -139,12 +188,70 @@ class TinyCodeAgent:
139
188
  final_config["authorized_imports"] = final_authorized_imports
140
189
  final_config["check_string_obfuscation"] = self.check_string_obfuscation
141
190
 
191
+ # Shell safety configuration (default to False for Modal)
192
+ bypass_shell_safety = config.get("bypass_shell_safety", False)
193
+ additional_safe_shell_commands = config.get("additional_safe_shell_commands", None)
194
+ additional_safe_control_operators = config.get("additional_safe_control_operators", None)
195
+
142
196
  return ModalProvider(
143
197
  log_manager=self.log_manager,
144
198
  code_tools=self.code_tools,
145
199
  local_execution=self.local_execution,
200
+ bypass_shell_safety=bypass_shell_safety,
201
+ additional_safe_shell_commands=additional_safe_shell_commands,
202
+ additional_safe_control_operators=additional_safe_control_operators,
146
203
  **final_config
147
204
  )
205
+ elif provider_type.lower() == "seatbelt":
206
+ # Check if seatbelt is supported on this system
207
+ if not SeatbeltProvider.is_supported():
208
+ raise ValueError("Seatbelt provider is not supported on this system. It requires macOS with sandbox-exec.")
209
+
210
+ # Seatbelt only works with local execution
211
+ if not self.local_execution:
212
+ raise ValueError("Seatbelt provider requires local execution mode. Please set local_execution=True.")
213
+
214
+ # Create a copy of the config without the parameters we'll pass directly
215
+ filtered_config = config.copy()
216
+ for key in ['seatbelt_profile', 'seatbelt_profile_path', 'python_env_path',
217
+ 'bypass_shell_safety', 'additional_safe_shell_commands',
218
+ 'additional_safe_control_operators', 'additional_read_dirs',
219
+ 'additional_write_dirs', 'environment_variables']:
220
+ if key in filtered_config:
221
+ filtered_config.pop(key)
222
+
223
+ # Get seatbelt profile configuration
224
+ seatbelt_profile = config.get("seatbelt_profile", None)
225
+ seatbelt_profile_path = config.get("seatbelt_profile_path", None)
226
+ python_env_path = config.get("python_env_path", None)
227
+
228
+ # Shell safety configuration (default to True for Seatbelt)
229
+ bypass_shell_safety = config.get("bypass_shell_safety", True)
230
+ additional_safe_shell_commands = config.get("additional_safe_shell_commands", None)
231
+ additional_safe_control_operators = config.get("additional_safe_control_operators", None)
232
+
233
+ # Additional directory access configuration
234
+ additional_read_dirs = config.get("additional_read_dirs", None)
235
+ additional_write_dirs = config.get("additional_write_dirs", None)
236
+
237
+ # Environment variables to make available in the sandbox
238
+ environment_variables = config.get("environment_variables", {})
239
+
240
+ # Create the seatbelt provider
241
+ return SeatbeltProvider(
242
+ log_manager=self.log_manager,
243
+ code_tools=self.code_tools,
244
+ seatbelt_profile=seatbelt_profile,
245
+ seatbelt_profile_path=seatbelt_profile_path,
246
+ python_env_path=python_env_path,
247
+ bypass_shell_safety=bypass_shell_safety,
248
+ additional_safe_shell_commands=additional_safe_shell_commands,
249
+ additional_safe_control_operators=additional_safe_control_operators,
250
+ additional_read_dirs=additional_read_dirs,
251
+ additional_write_dirs=additional_write_dirs,
252
+ environment_variables=environment_variables,
253
+ **filtered_config
254
+ )
148
255
  else:
149
256
  raise ValueError(f"Unsupported provider type: {provider_type}")
150
257
 
@@ -307,6 +414,24 @@ class TinyCodeAgent:
307
414
  # This ensures they stay in sync
308
415
  self.user_variables = self.code_provider.get_user_variables()
309
416
 
417
+ # Apply truncation if enabled
418
+ if self.truncation_config["enabled"] and "printed_output" in result:
419
+ truncated_output, is_truncated, original_tokens, original_lines = truncate_output(
420
+ result["printed_output"],
421
+ max_tokens=self.truncation_config["max_tokens"],
422
+ max_lines=self.truncation_config["max_lines"]
423
+ )
424
+
425
+ if is_truncated:
426
+ result["printed_output"] = format_truncation_message(
427
+ truncated_output,
428
+ is_truncated,
429
+ original_tokens,
430
+ original_lines,
431
+ self.truncation_config["max_lines"],
432
+ "python_output"
433
+ )
434
+
310
435
  return json.dumps(result)
311
436
  except Exception as e:
312
437
  print("!"*100)
@@ -348,7 +473,7 @@ class TinyCodeAgent:
348
473
  - You have the capability to call multiple tools in a single response. When multiple independent pieces of information are requested, batch your tool calls together for optimal performance.
349
474
  - You can specify an optional timeout in milliseconds (up to 600000ms / 10 minutes). If not specified, commands will timeout after 120000ms (2 minutes).
350
475
  - It is very helpful if you write a clear, concise description of what this command does in 5-10 words.
351
- - If the output exceeds 30000 characters, output will be truncated before being returned to you.
476
+ - If the output is too large, it will be truncated before being returned to you.
352
477
 
353
478
  - If you _still_ need to run `grep`, STOP. ALWAYS USE ripgrep at `rg` first, which all ${PRODUCT_NAME} users have pre-installed.
354
479
  " - When issuing multiple commands, use the ; or && operator to separate them. DO NOT use newlines (newlines are ok in quoted strings).\n" +
@@ -359,19 +484,66 @@ class TinyCodeAgent:
359
484
  <bad-example>
360
485
  cd /foo/bar && pytest tests
361
486
  </bad-example>
487
+
488
+ ## IMPORTANT: Bash Tool Usage
489
+
490
+ When using the bash tool, you MUST provide all required parameters:
491
+
492
+ **Correct Usage:**
493
+ ```
494
+ bash(
495
+ command=["ls", "-la"],
496
+ absolute_workdir="/path/to/directory",
497
+ description="List files in directory"
498
+ )
499
+ ```
500
+
501
+ **For creating files with content, use these safe patterns:**
502
+
503
+ 1. **Simple file creation:**
504
+ ```
505
+ bash(
506
+ command=["touch", "filename.txt"],
507
+ absolute_workdir="/working/directory",
508
+ description="Create empty file"
509
+ )
510
+ ```
511
+
512
+ 2. **Write content using cat and heredoc:**
513
+ ```
514
+ bash(
515
+ command=["sh", "-c", "cat > filename.txt << 'EOF'\nYour content here\nEOF"],
516
+ absolute_workdir="/working/directory",
517
+ description="Create file with content"
518
+ )
519
+ ```
520
+
521
+ 3. **Write content using echo:**
522
+ ```
523
+ bash(
524
+ command=["sh", "-c", "echo 'Your content' > filename.txt"],
525
+ absolute_workdir="/working/directory",
526
+ description="Write content to file"
527
+ )
528
+ ```
529
+
530
+ **Never:**
531
+ - Call bash() without all required parameters
532
+ - Use complex nested quotes without testing
533
+ - Try to create large files in a single command (break into parts)
362
534
 
363
535
  Args:
364
536
  command: list[str]: The shell command to execute as a list of strings. Example: ["ls", "-la"] or ["cat", "file.txt"]
365
537
 
366
538
  absolute_workdir: str: could be presented workdir in the system prompt or one of the subdirectories of the workdir. This is the only allowed path, and accessing else will result in an error.
367
539
  description: str: A clear, concise description of what this command does in 5-10 words.
368
- timeout: int: Maximum execution time in seconds (default: 30).
540
+ timeout: int: Maximum execution time in seconds (default: 60).
369
541
  Returns:
370
542
  Dictionary with stdout, stderr, and exit_code from the command execution.
371
543
  If the command is rejected for security reasons, stderr will contain the reason.
372
544
  The stdout will include information about which working directory was used.
373
545
  """))
374
- async def run_shell(command: List[str], absolute_workdir: str, description: str, timeout: int = 30) -> str:
546
+ async def run_shell(command: List[str], absolute_workdir: str, description: str, timeout: int = 60) -> str:
375
547
  """Execute shell commands securely using the configured provider."""
376
548
  try:
377
549
  # Use the default working directory if none is specified
@@ -393,6 +565,30 @@ class TinyCodeAgent:
393
565
  })
394
566
 
395
567
  result = await self.code_provider.execute_shell(command, timeout, effective_workdir)
568
+
569
+ # Apply truncation if enabled
570
+ if self.truncation_config["enabled"] and "stdout" in result and result["stdout"]:
571
+ truncated_output, is_truncated, original_tokens, original_lines = truncate_output(
572
+ result["stdout"],
573
+ max_tokens=self.truncation_config["max_tokens"],
574
+ max_lines=self.truncation_config["max_lines"]
575
+ )
576
+
577
+ if is_truncated:
578
+ result["stdout"] = format_truncation_message(
579
+ truncated_output,
580
+ is_truncated,
581
+ original_tokens,
582
+ original_lines,
583
+ self.truncation_config["max_lines"],
584
+ "bash_output"
585
+ )
586
+
587
+ # Create a git checkpoint if auto_git_checkpoint is enabled
588
+ if self.auto_git_checkpoint and result.get("exit_code", 1) == 0:
589
+ checkpoint_result = await self._create_git_checkpoint(command, description, effective_workdir)
590
+ self.log_manager.get_logger(__name__).info(f"Git checkpoint {effective_workdir} result: {checkpoint_result}")
591
+
396
592
  return json.dumps(result)
397
593
  except Exception as e:
398
594
  COLOR = {
@@ -407,6 +603,64 @@ class TinyCodeAgent:
407
603
  self.agent.add_tool(run_python)
408
604
  self.agent.add_tool(run_shell)
409
605
 
606
+ async def _create_git_checkpoint(self, command: List[str], description: str, workdir: str) -> Dict[str, Any]:
607
+ """
608
+ Create a git checkpoint after command execution.
609
+
610
+ Args:
611
+ command: The command that was executed
612
+ description: Description of the command
613
+ workdir: Working directory where the command was executed
614
+
615
+ Returns:
616
+ Dictionary with stdout and stderr from the git operations
617
+ """
618
+ try:
619
+ # Format the command for the commit message
620
+ cmd_str = " ".join(command)
621
+
622
+ # Check if there are changes to commit
623
+ git_check_cmd = ["bash", "-c", "if ! git diff-index --quiet HEAD --; then echo 'changes_exist'; else echo 'no_changes'; fi"]
624
+ check_result = await self.code_provider.execute_shell(git_check_cmd, 10, workdir)
625
+
626
+ # If no changes or check failed, return early
627
+ if check_result.get("exit_code", 1) != 0 or "no_changes" in check_result.get("stdout", ""):
628
+ return {"stdout": "No changes detected, skipping git checkpoint", "stderr": ""}
629
+
630
+ # Stage all changes
631
+ git_add_cmd = ["git", "add", "-A"]
632
+ add_result = await self.code_provider.execute_shell(git_add_cmd, 30, workdir)
633
+
634
+ if add_result.get("exit_code", 1) != 0:
635
+ return {
636
+ "stdout": "",
637
+ "stderr": f"Failed to stage changes: {add_result.get('stderr', '')}"
638
+ }
639
+
640
+ # Create commit with command description and timestamp
641
+ timestamp = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')
642
+ commit_msg = f"Checkpoint: {description} @ {timestamp}\n\nCommand: {cmd_str}"
643
+ git_commit_cmd = ["git", "commit", "-m", commit_msg, "--no-gpg-sign"]
644
+ commit_result = await self.code_provider.execute_shell(git_commit_cmd, 30, workdir)
645
+
646
+ if commit_result.get("exit_code", 1) != 0:
647
+ return {
648
+ "stdout": "",
649
+ "stderr": f"Failed to create commit: {commit_result.get('stderr', '')}"
650
+ }
651
+
652
+ # Get the first line of the commit message without using split with \n in f-string
653
+ first_line = commit_msg.split("\n")[0]
654
+ return {
655
+ "stdout": f"✓ Git checkpoint created: {first_line}",
656
+ "stderr": ""
657
+ }
658
+ except Exception as e:
659
+ return {
660
+ "stdout": "",
661
+ "stderr": f"Error creating git checkpoint: {str(e)}"
662
+ }
663
+
410
664
  def set_default_workdir(self, workdir: str, create_if_not_exists: bool = False):
411
665
  """
412
666
  Set the default working directory for shell commands.
@@ -676,6 +930,17 @@ class TinyCodeAgent:
676
930
  """
677
931
  return self.authorized_imports.copy()
678
932
 
933
+ @classmethod
934
+ def is_seatbelt_supported(cls) -> bool:
935
+ """
936
+ Check if the seatbelt provider is supported on this system.
937
+
938
+ Returns:
939
+ True if seatbelt is supported (macOS with sandbox-exec), False otherwise
940
+ """
941
+ from .providers.seatbelt_provider import SeatbeltProvider
942
+ return SeatbeltProvider.is_supported()
943
+
679
944
  def remove_authorized_import(self, import_name: str):
680
945
  """
681
946
  Remove an authorized import.
@@ -794,6 +1059,119 @@ class TinyCodeAgent:
794
1059
  else:
795
1060
  self.log_manager.get_logger(__name__).warning(f"Unknown UI type: {ui_type}. No UI callback will be added.")
796
1061
 
1062
+ def set_truncation_config(self, config: Dict[str, Any]):
1063
+ """
1064
+ Set the truncation configuration.
1065
+
1066
+ Args:
1067
+ config: Dictionary containing truncation configuration options:
1068
+ - max_tokens: Maximum number of tokens to keep in output
1069
+ - max_lines: Maximum number of lines to keep in output
1070
+ - enabled: Whether truncation is enabled
1071
+ """
1072
+ self.truncation_config.update(config)
1073
+
1074
+ def get_truncation_config(self) -> Dict[str, Any]:
1075
+ """
1076
+ Get the current truncation configuration.
1077
+
1078
+ Returns:
1079
+ Dictionary containing truncation configuration
1080
+ """
1081
+ return self.truncation_config.copy()
1082
+
1083
+ def enable_truncation(self, enabled: bool = True):
1084
+ """
1085
+ Enable or disable output truncation.
1086
+
1087
+ Args:
1088
+ enabled: Whether to enable truncation
1089
+ """
1090
+ self.truncation_config["enabled"] = enabled
1091
+
1092
+ def enable_auto_git_checkpoint(self, enabled: bool = True):
1093
+ """
1094
+ Enable or disable automatic git checkpoint creation after successful shell commands.
1095
+
1096
+ Args:
1097
+ enabled: If True, automatically create git checkpoints. If False, do not create them.
1098
+ """
1099
+ self.auto_git_checkpoint = enabled
1100
+
1101
+ def get_auto_git_checkpoint_status(self) -> bool:
1102
+ """
1103
+ Get the current status of auto_git_checkpoint.
1104
+
1105
+ Returns:
1106
+ True if auto_git_checkpoint is enabled, False otherwise.
1107
+ """
1108
+ return self.auto_git_checkpoint
1109
+
1110
+ def set_environment_variables(self, env_vars: Dict[str, str]):
1111
+ """
1112
+ Set environment variables for the code execution provider.
1113
+ Currently only supported for SeatbeltProvider.
1114
+
1115
+ Args:
1116
+ env_vars: Dictionary of environment variable name -> value pairs
1117
+
1118
+ Raises:
1119
+ AttributeError: If the provider doesn't support environment variables
1120
+ """
1121
+ if hasattr(self.code_provider, 'set_environment_variables'):
1122
+ self.code_provider.set_environment_variables(env_vars)
1123
+ else:
1124
+ raise AttributeError(f"Provider {self.provider} does not support environment variables")
1125
+
1126
+ def add_environment_variable(self, name: str, value: str):
1127
+ """
1128
+ Add a single environment variable for the code execution provider.
1129
+ Currently only supported for SeatbeltProvider.
1130
+
1131
+ Args:
1132
+ name: Environment variable name
1133
+ value: Environment variable value
1134
+
1135
+ Raises:
1136
+ AttributeError: If the provider doesn't support environment variables
1137
+ """
1138
+ if hasattr(self.code_provider, 'add_environment_variable'):
1139
+ self.code_provider.add_environment_variable(name, value)
1140
+ else:
1141
+ raise AttributeError(f"Provider {self.provider} does not support environment variables")
1142
+
1143
+ def remove_environment_variable(self, name: str):
1144
+ """
1145
+ Remove an environment variable from the code execution provider.
1146
+ Currently only supported for SeatbeltProvider.
1147
+
1148
+ Args:
1149
+ name: Environment variable name to remove
1150
+
1151
+ Raises:
1152
+ AttributeError: If the provider doesn't support environment variables
1153
+ """
1154
+ if hasattr(self.code_provider, 'remove_environment_variable'):
1155
+ self.code_provider.remove_environment_variable(name)
1156
+ else:
1157
+ raise AttributeError(f"Provider {self.provider} does not support environment variables")
1158
+
1159
+ def get_environment_variables(self) -> Dict[str, str]:
1160
+ """
1161
+ Get a copy of current environment variables from the code execution provider.
1162
+ Currently only supported for SeatbeltProvider.
1163
+
1164
+ Returns:
1165
+ Dictionary of current environment variables
1166
+
1167
+ Raises:
1168
+ AttributeError: If the provider doesn't support environment variables
1169
+ """
1170
+ if hasattr(self.code_provider, 'get_environment_variables'):
1171
+ return self.code_provider.get_environment_variables()
1172
+ else:
1173
+ raise AttributeError(f"Provider {self.provider} does not support environment variables")
1174
+
797
1175
 
798
1176
  # Example usage demonstrating both LLM tools and code tools
799
1177
  async def run_example():
@@ -836,7 +1214,12 @@ async def run_example():
836
1214
  authorized_imports=["tinyagent", "gradio", "requests", "numpy", "pandas"], # Explicitly specify authorized imports
837
1215
  local_execution=False, # Remote execution via Modal (default)
838
1216
  check_string_obfuscation=True,
839
- default_workdir=os.path.join(os.getcwd(), "examples") # Set a default working directory for shell commands
1217
+ default_workdir=os.path.join(os.getcwd(), "examples"), # Set a default working directory for shell commands
1218
+ truncation_config={
1219
+ "max_tokens": 3000,
1220
+ "max_lines": 250,
1221
+ "enabled": True
1222
+ }
840
1223
  )
841
1224
 
842
1225
  # Connect to MCP servers
@@ -961,6 +1344,292 @@ async def run_example():
961
1344
  print("Shell Execution with Explicit Working Directory:")
962
1345
  print(response_shell_explicit)
963
1346
 
1347
+ # Test truncation functionality
1348
+ print("\n" + "="*80)
1349
+ print("✂️ Testing output truncation")
1350
+
1351
+ # Configure truncation with smaller limits for testing
1352
+ agent_remote.set_truncation_config({
1353
+ "max_tokens": 100, # Very small limit for testing
1354
+ "max_lines": 5 # Very small limit for testing
1355
+ })
1356
+
1357
+ # Generate a large output to test truncation
1358
+ large_output_prompt = """
1359
+ Generate a large output by printing a lot of text. Create a Python script that:
1360
+ 1. Prints numbers from 1 to 1000
1361
+ 2. For each number, also print its square and cube
1362
+ 3. Add random text for each line to make it longer
1363
+ """
1364
+
1365
+ response_truncated = await agent_remote.run(large_output_prompt)
1366
+ print("Truncated Output Response:")
1367
+ print(response_truncated)
1368
+
1369
+ # Test disabling truncation
1370
+ print("\n" + "="*80)
1371
+ print("🔄 Testing with truncation disabled")
1372
+
1373
+ agent_remote.enable_truncation(False)
1374
+ response_untruncated = await agent_remote.run("Run the same script again but limit to 20 numbers")
1375
+ print("Untruncated Output Response:")
1376
+ print(response_untruncated)
1377
+
1378
+ # Test git checkpoint functionality
1379
+ print("\n" + "="*80)
1380
+ print("🔄 Testing git checkpoint functionality")
1381
+
1382
+ # Enable git checkpoints
1383
+ agent_remote.enable_auto_git_checkpoint(True)
1384
+ print(f"Auto Git Checkpoint enabled: {agent_remote.get_auto_git_checkpoint_status()}")
1385
+
1386
+ # Create a test file to demonstrate git checkpoint
1387
+ git_test_prompt = """
1388
+ Create a new file called test_file.txt with some content, then modify it, and observe
1389
+ that git checkpoints are created automatically after each change.
1390
+ """
1391
+
1392
+ git_response = await agent_remote.run(git_test_prompt)
1393
+ print("Git Checkpoint Response:")
1394
+ print(git_response)
1395
+
1396
+ # Disable git checkpoints
1397
+ agent_remote.enable_auto_git_checkpoint(False)
1398
+ print(f"Auto Git Checkpoint disabled: {agent_remote.get_auto_git_checkpoint_status()}")
1399
+
1400
+ # Test seatbelt provider if supported
1401
+ if TinyCodeAgent.is_seatbelt_supported():
1402
+ print("\n" + "="*80)
1403
+ print("🔒 Testing TinyCodeAgent with SEATBELT provider (sandboxed execution)")
1404
+
1405
+ # Create a test directory for read/write access
1406
+ test_read_dir = os.path.join(os.getcwd(), "test_read_dir")
1407
+ test_write_dir = os.path.join(os.getcwd(), "test_write_dir")
1408
+
1409
+ # Create directories if they don't exist
1410
+ os.makedirs(test_read_dir, exist_ok=True)
1411
+ os.makedirs(test_write_dir, exist_ok=True)
1412
+
1413
+ # Create a test file in the read directory
1414
+ with open(os.path.join(test_read_dir, "test.txt"), "w") as f:
1415
+ f.write("This is a test file for reading")
1416
+
1417
+ # Create a simple seatbelt profile
1418
+ seatbelt_profile = """(version 1)
1419
+
1420
+ ; Default to deny everything
1421
+ (deny default)
1422
+
1423
+ ; Allow network connections with proper DNS resolution
1424
+ (allow network*)
1425
+ (allow network-outbound)
1426
+ (allow mach-lookup)
1427
+
1428
+ ; Allow process execution
1429
+ (allow process-exec)
1430
+ (allow process-fork)
1431
+ (allow signal (target self))
1432
+
1433
+ ; Restrict file read to current path and system files
1434
+ (deny file-read* (subpath "/Users"))
1435
+ (allow file-read*
1436
+ (subpath "{os.getcwd()}")
1437
+ (subpath "/usr")
1438
+ (subpath "/System")
1439
+ (subpath "/Library")
1440
+ (subpath "/bin")
1441
+ (subpath "/sbin")
1442
+ (subpath "/opt")
1443
+ (subpath "/private/tmp")
1444
+ (subpath "/private/var/tmp")
1445
+ (subpath "/dev")
1446
+ (subpath "/etc")
1447
+ (literal "/")
1448
+ (literal "/."))
1449
+
1450
+ ; Allow write access to specified folder and temp directories
1451
+ (deny file-write* (subpath "/"))
1452
+ (allow file-write*
1453
+ (subpath "{os.getcwd()}")
1454
+ (subpath "/private/tmp")
1455
+ (subpath "/private/var/tmp")
1456
+ (subpath "/dev"))
1457
+
1458
+ ; Allow standard device operations
1459
+ (allow file-write-data
1460
+ (literal "/dev/null")
1461
+ (literal "/dev/dtracehelper")
1462
+ (literal "/dev/tty")
1463
+ (literal "/dev/stdout")
1464
+ (literal "/dev/stderr"))
1465
+
1466
+ ; Allow iokit operations needed for system functions
1467
+ (allow iokit-open)
1468
+
1469
+ ; Allow shared memory operations
1470
+ (allow ipc-posix-shm)
1471
+
1472
+ ; Allow basic system operations
1473
+ (allow file-read-metadata)
1474
+ (allow process-info-pidinfo)
1475
+ (allow process-info-setcontrol)
1476
+ """
1477
+
1478
+ # Create TinyCodeAgent with seatbelt provider
1479
+ agent_seatbelt = TinyCodeAgent(
1480
+ model="gpt-4.1-mini",
1481
+ tools=[search_web], # LLM tools
1482
+ code_tools=[data_processor], # Code tools
1483
+ user_variables={
1484
+ "sample_data": [1, 2, 3, 4, 5, 10, 15, 20]
1485
+ },
1486
+ provider="seatbelt", # Use seatbelt provider
1487
+ provider_config={
1488
+ "seatbelt_profile": seatbelt_profile,
1489
+ # Alternatively, you can specify a path to a seatbelt profile file:
1490
+ # "seatbelt_profile_path": "/path/to/seatbelt.sb",
1491
+ # "python_env_path": "/path/to/python/env", # Optional path to Python environment
1492
+
1493
+ # Specify additional directories for read/write access
1494
+ "additional_read_dirs": [test_read_dir],
1495
+ "additional_write_dirs": [test_write_dir],
1496
+
1497
+ # Allow git commands
1498
+ "bypass_shell_safety": True,
1499
+ "additional_safe_shell_commands": ["git"],
1500
+
1501
+ # Environment variables to make available in the sandbox
1502
+ "environment_variables": {
1503
+ "TEST_READ_DIR": test_read_dir,
1504
+ "TEST_WRITE_DIR": test_write_dir,
1505
+ "PROJECT_NAME": "TinyAgent Seatbelt Demo",
1506
+ "BUILD_VERSION": "1.0.0"
1507
+ }
1508
+ },
1509
+ local_execution=True, # Required for seatbelt
1510
+ check_string_obfuscation=True,
1511
+ truncation_config={
1512
+ "max_tokens": 500,
1513
+ "max_lines": 20,
1514
+ "enabled": True
1515
+ }
1516
+ )
1517
+
1518
+ # Connect to MCP servers
1519
+ await agent_seatbelt.connect_to_server("npx", ["-y", "@openbnb/mcp-server-airbnb", "--ignore-robots-txt"])
1520
+ await agent_seatbelt.connect_to_server("npx", ["-y", "@modelcontextprotocol/server-sequential-thinking"])
1521
+
1522
+ # Test the seatbelt agent
1523
+ response_seatbelt = await agent_seatbelt.run("""
1524
+ I have some sample data. Please use the data_processor tool in Python to analyze my sample_data
1525
+ and show me the results.
1526
+ """)
1527
+
1528
+ print("Seatbelt Agent Response:")
1529
+ print(response_seatbelt)
1530
+
1531
+ # Test shell execution in sandbox
1532
+ shell_prompt_sandbox = "Run 'ls -la' to list files in the current directory."
1533
+
1534
+ response_shell_sandbox = await agent_seatbelt.run(shell_prompt_sandbox)
1535
+ print("Shell Execution in Sandbox:")
1536
+ print(response_shell_sandbox)
1537
+
1538
+ # Test reading from the additional read directory
1539
+ read_prompt = f"Read the contents of the file in the test_read_dir directory."
1540
+
1541
+ response_read = await agent_seatbelt.run(read_prompt)
1542
+ print("Reading from Additional Read Directory:")
1543
+ print(response_read)
1544
+
1545
+ # Test writing to the additional write directory
1546
+ write_prompt = f"Write a file called 'output.txt' with the text 'Hello from sandbox!' in the test_write_dir directory."
1547
+
1548
+ response_write = await agent_seatbelt.run(write_prompt)
1549
+ print("Writing to Additional Write Directory:")
1550
+ print(response_write)
1551
+
1552
+ # Test environment variables
1553
+ print("\n" + "="*80)
1554
+ print("🔧 Testing environment variables functionality")
1555
+
1556
+ # Add additional environment variables dynamically
1557
+ agent_seatbelt.add_environment_variable("CUSTOM_VAR", "custom_value")
1558
+ agent_seatbelt.add_environment_variable("DEBUG_MODE", "true")
1559
+
1560
+ # Get and display current environment variables
1561
+ current_env_vars = agent_seatbelt.get_environment_variables()
1562
+ print(f"Current environment variables: {list(current_env_vars.keys())}")
1563
+
1564
+ # Test accessing environment variables in Python and shell
1565
+ env_test_prompt = """
1566
+ Test the environment variables we set:
1567
+ 1. In Python, use os.environ to check for CUSTOM_VAR and DEBUG_MODE
1568
+ 2. In a shell command, use 'echo $CUSTOM_VAR' and 'echo $DEBUG_MODE'
1569
+ 3. Also check the TEST_READ_DIR and TEST_WRITE_DIR variables that were set during initialization
1570
+ 4. Show all environment variables that start with 'TEST_' or 'CUSTOM_' or 'DEBUG_'
1571
+ """
1572
+
1573
+ response_env_test = await agent_seatbelt.run(env_test_prompt)
1574
+ print("Environment Variables Test:")
1575
+ print(response_env_test)
1576
+
1577
+ # Update environment variables
1578
+ agent_seatbelt.set_environment_variables({
1579
+ "CUSTOM_VAR": "updated_value",
1580
+ "NEW_VAR": "new_value",
1581
+ "API_KEY": "test_api_key_123"
1582
+ })
1583
+
1584
+ # Test updated environment variables
1585
+ updated_env_test_prompt = """
1586
+ Test the updated environment variables:
1587
+ 1. Check that CUSTOM_VAR now has the value 'updated_value'
1588
+ 2. Check that NEW_VAR is available with value 'new_value'
1589
+ 3. Check that API_KEY is available with value 'test_api_key_123'
1590
+ 4. Verify that DEBUG_MODE is no longer available (should have been removed by set operation)
1591
+ """
1592
+
1593
+ response_updated_env = await agent_seatbelt.run(updated_env_test_prompt)
1594
+ print("Updated Environment Variables Test:")
1595
+ print(response_updated_env)
1596
+
1597
+ # Remove a specific environment variable
1598
+ agent_seatbelt.remove_environment_variable("API_KEY")
1599
+
1600
+ # Test that the removed variable is no longer available
1601
+ removed_env_test_prompt = """
1602
+ Test that API_KEY environment variable has been removed:
1603
+ 1. Try to access API_KEY in Python - it should not be available
1604
+ 2. Use shell command 'echo $API_KEY' - it should be empty
1605
+ 3. List all current environment variables that start with 'CUSTOM_' or 'NEW_'
1606
+ """
1607
+
1608
+ response_removed_env = await agent_seatbelt.run(removed_env_test_prompt)
1609
+ print("Removed Environment Variable Test:")
1610
+ print(response_removed_env)
1611
+
1612
+ # Test git commands with the custom configuration
1613
+ git_prompt = "Run 'git status' to show the current git status."
1614
+
1615
+ response_git = await agent_seatbelt.run(git_prompt)
1616
+ print("Git Command Execution:")
1617
+ print(response_git)
1618
+
1619
+ # Clean up test directories
1620
+ import shutil
1621
+ try:
1622
+ shutil.rmtree(test_read_dir)
1623
+ shutil.rmtree(test_write_dir)
1624
+ print("Cleaned up test directories")
1625
+ except Exception as e:
1626
+ print(f"Error cleaning up test directories: {str(e)}")
1627
+
1628
+ await agent_seatbelt.close()
1629
+ else:
1630
+ print("\n" + "="*80)
1631
+ print("⚠️ Seatbelt provider is not supported on this system. Skipping seatbelt tests.")
1632
+
964
1633
  await agent_remote.close()
965
1634
  await agent_local.close()
966
1635