Refactor Terminal and AIAgent cleanup

This commit is contained in:
teknium1 2026-02-21 22:31:43 -08:00
parent 9018e9dd70
commit 9123cfb5dd
17 changed files with 1842 additions and 976 deletions

View file

@ -75,125 +75,8 @@ if mini_swe_path.exists():
# Custom Singularity Environment with more space
# =============================================================================
def _get_scratch_dir() -> Path:
"""Get the best directory for Singularity sandboxes - prefers /scratch if available."""
# Check for configurable scratch directory first (highest priority)
custom_scratch = os.getenv("TERMINAL_SCRATCH_DIR")
if custom_scratch:
scratch_path = Path(custom_scratch)
scratch_path.mkdir(parents=True, exist_ok=True)
return scratch_path
# Check for /scratch (common on HPC clusters, especially GPU nodes)
scratch = Path("/scratch")
if scratch.exists() and os.access(scratch, os.W_OK):
# Create user-specific subdirectory
user_scratch = scratch / os.getenv("USER", "hermes") / "hermes-agent"
user_scratch.mkdir(parents=True, exist_ok=True)
logger.info("Using /scratch for sandboxes: %s", user_scratch)
return user_scratch
# Fall back to /tmp (only relevant for Singularity/HPC sandboxes)
logger.debug("/scratch not available, using /tmp for sandboxes")
return Path(tempfile.gettempdir())
def _get_apptainer_cache_dir() -> Path:
"""Get the Apptainer cache directory for SIF images."""
# Check for APPTAINER_CACHEDIR env var
cache_dir = os.getenv("APPTAINER_CACHEDIR")
if cache_dir:
cache_path = Path(cache_dir)
cache_path.mkdir(parents=True, exist_ok=True)
return cache_path
# Use user-specific subdirectory in scratch for cache
scratch = _get_scratch_dir()
cache_path = scratch / ".apptainer"
cache_path.mkdir(parents=True, exist_ok=True)
return cache_path
# Lock for SIF building to prevent race conditions
_sif_build_lock = threading.Lock()
def _get_or_build_sif(image: str, executable: str = "apptainer") -> str:
"""
Get or build a SIF image from a docker:// URL.
If the image is already a .sif file, returns it as-is.
If the image is a docker:// URL, checks for cached SIF and builds if needed.
Args:
image: Image path (docker://... URL or .sif path)
executable: apptainer or singularity
Returns:
Path to SIF file, or original image if not a docker:// URL
"""
# If already a .sif file, use it directly
if image.endswith('.sif') and Path(image).exists():
return image
# If not a docker:// URL, return as-is (could be a local sandbox or other format)
if not image.startswith('docker://'):
return image
# Generate SIF filename from docker image name
# docker://nikolaik/python-nodejs:python3.11-nodejs20 -> python-nodejs-python3.11-nodejs20.sif
image_name = image.replace('docker://', '').replace('/', '-').replace(':', '-')
cache_dir = _get_apptainer_cache_dir()
sif_path = cache_dir / f"{image_name}.sif"
# Check if SIF already exists
if sif_path.exists():
return str(sif_path)
# Build SIF with lock to prevent multiple workers building simultaneously
with _sif_build_lock:
# Double-check after acquiring lock (another thread may have built it)
if sif_path.exists():
return str(sif_path)
logger.info("Building SIF image (one-time setup)...")
logger.info(" Source: %s", image)
logger.info(" Target: %s", sif_path)
# Ensure tmp directory exists for build
tmp_dir = cache_dir / "tmp"
tmp_dir.mkdir(parents=True, exist_ok=True)
# Set APPTAINER_TMPDIR for the build
env = os.environ.copy()
env["APPTAINER_TMPDIR"] = str(tmp_dir)
env["APPTAINER_CACHEDIR"] = str(cache_dir)
try:
result = subprocess.run(
[executable, "build", str(sif_path), image],
capture_output=True,
text=True,
timeout=600, # 10 min timeout for pulling and building
env=env
)
if result.returncode != 0:
logger.warning("SIF build failed, falling back to docker:// URL")
logger.warning(" Error: %s", result.stderr[:500])
return image
logger.info("SIF image built successfully")
return str(sif_path)
except subprocess.TimeoutExpired:
logger.warning("SIF build timed out, falling back to docker:// URL")
# Clean up partial file
if sif_path.exists():
sif_path.unlink()
return image
except Exception as e:
logger.warning("SIF build error: %s, falling back to docker:// URL", e)
return image
# Singularity helpers (scratch dir, SIF cache) now live in tools/environments/singularity.py
from tools.environments.singularity import _get_scratch_dir
# Disk usage warning threshold (in GB)
@ -255,234 +138,19 @@ def set_approval_callback(cb):
# Dangerous Command Approval System
# =============================================================================
from tools import approval as _approval
# Dangerous command patterns (regex, description)
DANGEROUS_PATTERNS = [
(r'\brm\s+(-[^\s]*\s+)*/', "delete in root path"),
(r'\brm\s+(-[^\s]*)?r', "recursive delete"),
(r'\brm\s+--recursive\b', "recursive delete (long flag)"),
(r'\bchmod\s+(-[^\s]*\s+)*777\b', "world-writable permissions"),
(r'\bchmod\s+--recursive\b.*777', "recursive world-writable (long flag)"),
(r'\bchown\s+(-[^\s]*)?R\s+root', "recursive chown to root"),
(r'\bchown\s+--recursive\b.*root', "recursive chown to root (long flag)"),
(r'\bmkfs\b', "format filesystem"),
(r'\bdd\s+.*if=', "disk copy"),
(r'>\s*/dev/sd', "write to block device"),
(r'\bDROP\s+(TABLE|DATABASE)\b', "SQL DROP"),
(r'\bDELETE\s+FROM\b(?!.*\bWHERE\b)', "SQL DELETE without WHERE"),
(r'\bTRUNCATE\s+(TABLE)?\s*\w', "SQL TRUNCATE"),
(r'>\s*/etc/', "overwrite system config"),
(r'\bsystemctl\s+(stop|disable|mask)\b', "stop/disable system service"),
(r'\bkill\s+-9\s+-1\b', "kill all processes"),
(r'\bpkill\s+-9\b', "force kill processes"),
(r':()\s*{\s*:\s*\|\s*:&\s*}\s*;:', "fork bomb"),
# Indirect execution via command launchers
(r'\b(bash|sh|zsh)\s+-c\s+', "shell command via -c flag"),
(r'\b(python[23]?|perl|ruby|node)\s+-[ec]\s+', "script execution via -e/-c flag"),
# Pipe-to-shell (remote code execution)
(r'\b(curl|wget)\b.*\|\s*(ba)?sh\b', "pipe remote content to shell"),
# Destructive find/xargs patterns
(r'\bxargs\s+.*\brm\b', "xargs with rm"),
(r'\bfind\b.*-exec\s+rm\b', "find -exec rm"),
(r'\bfind\b.*-delete\b', "find -delete"),
]
def _load_permanent_allowlist() -> set:
"""Load permanently allowed command patterns from config.
Also syncs them into the approval module so is_approved() works for
patterns that were added via 'always' in a previous session.
"""
try:
from hermes_cli.config import load_config
config = load_config()
patterns = set(config.get("command_allowlist", []) or [])
if patterns:
_approval.load_permanent(patterns)
return patterns
except Exception:
return set()
def _save_permanent_allowlist(patterns: set):
"""Save permanently allowed command patterns to config."""
try:
from hermes_cli.config import load_config, save_config
config = load_config()
config["command_allowlist"] = list(patterns)
save_config(config)
except Exception as e:
logger.warning("Could not save allowlist: %s", e)
def _detect_dangerous_command(command: str) -> tuple:
"""
Check if command matches any dangerous patterns.
Returns:
(is_dangerous, pattern_key, description) or (False, None, None)
"""
import re
command_lower = command.lower()
for pattern, description in DANGEROUS_PATTERNS:
if re.search(pattern, command_lower, re.IGNORECASE):
# Use a simplified pattern key for caching (first word + key chars)
pattern_key = pattern.split(r'\b')[1] if r'\b' in pattern else pattern[:20]
return (True, pattern_key, description)
return (False, None, None)
def _is_command_approved(pattern_key: str) -> bool:
"""Check if a pattern is approved (session or permanent)."""
session_key = os.getenv("HERMES_SESSION_KEY", "default")
return _approval.is_approved(session_key, pattern_key)
def _prompt_dangerous_approval(command: str, description: str, timeout_seconds: int = 60) -> str:
"""
Prompt user to approve a dangerous command (CLI only).
If an _approval_callback is registered (by the CLI), delegates to it so the
prompt integrates with prompt_toolkit's UI. Otherwise falls back to the
raw input() approach (works outside the TUI, e.g. tests).
Returns: 'once', 'session', 'always', or 'deny'
"""
import sys
import threading
# Use the registered callback when available (prompt_toolkit-compatible)
if _approval_callback is not None:
try:
return _approval_callback(command, description)
except Exception:
return "deny"
# Pause spinner if one is running
os.environ["HERMES_SPINNER_PAUSE"] = "1"
try:
print()
print(f" ⚠️ DANGEROUS COMMAND: {description}")
print(f" {command[:80]}{'...' if len(command) > 80 else ''}")
print()
print(f" [o]nce | [s]ession | [a]lways | [d]eny")
print()
sys.stdout.flush()
result = {"choice": ""}
def get_input():
try:
result["choice"] = input(" Choice [o/s/a/D]: ").strip().lower()
except (EOFError, OSError):
result["choice"] = ""
thread = threading.Thread(target=get_input, daemon=True)
thread.start()
thread.join(timeout=timeout_seconds)
if thread.is_alive():
print("\n ⏱ Timeout - denying command")
return "deny"
choice = result["choice"]
if choice in ('o', 'once'):
print(" ✓ Allowed once")
return "once"
elif choice in ('s', 'session'):
print(" ✓ Allowed for this session")
return "session"
elif choice in ('a', 'always'):
print(" ✓ Added to permanent allowlist")
return "always"
else:
print(" ✗ Denied")
return "deny"
except (EOFError, KeyboardInterrupt):
print("\n ✗ Cancelled")
return "deny"
finally:
if "HERMES_SPINNER_PAUSE" in os.environ:
del os.environ["HERMES_SPINNER_PAUSE"]
print()
sys.stdout.flush()
# Dangerous command detection + approval now consolidated in tools/approval.py
from tools.approval import (
detect_dangerous_command as _detect_dangerous_command,
check_dangerous_command as _check_dangerous_command_impl,
load_permanent_allowlist as _load_permanent_allowlist,
DANGEROUS_PATTERNS,
)
def _check_dangerous_command(command: str, env_type: str) -> dict:
"""
Check if command is dangerous and handle approval.
Only applies to local/ssh backends in interactive contexts.
Args:
command: The command to check
env_type: The terminal backend type
Returns:
{"approved": True/False, "message": str or None}
"""
# Skip check for isolated environments (containers are disposable)
if env_type in ("docker", "singularity", "modal"):
return {"approved": True, "message": None}
# Detect dangerous command
is_dangerous, pattern_key, description = _detect_dangerous_command(command)
if not is_dangerous:
return {"approved": True, "message": None}
# Check if already approved
if _is_command_approved(pattern_key):
return {"approved": True, "message": None}
# Check context - only prompt in interactive modes
is_cli = os.getenv("HERMES_INTERACTIVE")
is_gateway = os.getenv("HERMES_GATEWAY_SESSION")
if not is_cli and not is_gateway:
# Programmatic use - allow (user opted into local backend)
return {"approved": True, "message": None}
if is_gateway or os.getenv("HERMES_EXEC_ASK"):
# Messaging context - return approval_required so the gateway can
# prompt the user interactively instead of just blocking
session_key = os.getenv("HERMES_SESSION_KEY", "default")
_approval.submit_pending(session_key, {
"command": command,
"pattern_key": pattern_key,
"description": description,
})
return {
"approved": False,
"pattern_key": pattern_key,
"status": "approval_required",
"command": command,
"description": description,
"message": f"⚠️ This command is potentially dangerous ({description}). Asking the user for approval..."
}
# CLI context - prompt user
choice = _prompt_dangerous_approval(command, description)
if choice == "deny":
return {"approved": False, "message": "BLOCKED: User denied this potentially dangerous command. Do NOT retry this command - the user has explicitly rejected it."}
session_key = os.getenv("HERMES_SESSION_KEY", "default")
if choice == "session":
_approval.approve_session(session_key, pattern_key)
elif choice == "always":
_approval.approve_session(session_key, pattern_key)
_approval.approve_permanent(pattern_key)
_save_permanent_allowlist(_load_permanent_allowlist() | {pattern_key})
return {"approved": True, "message": None}
"""Delegate to the consolidated approval module, passing the CLI callback."""
return _check_dangerous_command_impl(command, env_type,
approval_callback=_approval_callback)
def _handle_sudo_failure(output: str, env_type: str) -> str:
@ -671,569 +339,12 @@ def _transform_sudo_command(command: str) -> str:
return re.sub(r'\bsudo\b', replace_sudo, command)
class _LocalEnvironment:
"""
Local execution environment with sudo support and non-blocking stdin.
Features:
- Uses stdin=DEVNULL to prevent hanging on interactive prompts (sudo, etc.)
- Optional SUDO_PASSWORD support: if set, transforms `sudo` commands to use `sudo -S`
- Graceful failure: sudo commands fail fast with clear error if no password configured
Environment variables:
- SUDO_PASSWORD: If set, enables sudo commands by piping password via `sudo -S`
"""
def __init__(self, cwd: str = "", timeout: int = 60, env: dict = None):
self.cwd = cwd or os.getcwd()
self.timeout = timeout
self.env = env or {}
def execute(self, command: str, cwd: str = "", *, timeout: int | None = None,
stdin_data: str | None = None) -> dict:
"""
Execute a command locally with sudo support.
Uses Popen + polling so the global interrupt event can kill the
process early when the user sends a new message, instead of
blocking for the full timeout.
A background reader thread drains stdout continuously to prevent
pipe buffer deadlocks. Without this, commands producing >64KB of
output would block (Linux pipe buffer = 64KB) while the poll loop
waits for the process to finish a classic deadlock.
Args:
stdin_data: If provided, piped to the process's stdin. This
bypasses shell ARG_MAX limits for large content.
"""
work_dir = cwd or self.cwd or os.getcwd()
effective_timeout = timeout or self.timeout
# Transform sudo commands if SUDO_PASSWORD is available
exec_command = _transform_sudo_command(command)
try:
proc = subprocess.Popen(
exec_command,
shell=True,
text=True,
cwd=work_dir,
env=os.environ | self.env,
encoding="utf-8",
errors="replace",
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
stdin=subprocess.PIPE if stdin_data is not None else subprocess.DEVNULL,
# Start in a new process group so we can kill the whole tree
preexec_fn=os.setsid,
)
# Pipe stdin_data in a background thread to avoid deadlock
# (large writes can block if the pipe buffer fills before the
# process drains it).
if stdin_data is not None:
def _write_stdin():
try:
proc.stdin.write(stdin_data)
proc.stdin.close()
except (BrokenPipeError, OSError):
pass
stdin_writer = threading.Thread(target=_write_stdin, daemon=True)
stdin_writer.start()
# Drain stdout in a background thread to prevent pipe buffer
# deadlocks. The OS pipe buffer is 64KB on Linux; if the child
# writes more than that before anyone reads, it blocks forever.
_output_chunks: list[str] = []
def _drain_stdout():
try:
for line in proc.stdout:
_output_chunks.append(line)
except ValueError:
pass # stdout closed during interrupt/timeout
finally:
try:
proc.stdout.close()
except Exception:
pass
reader = threading.Thread(target=_drain_stdout, daemon=True)
reader.start()
deadline = time.monotonic() + effective_timeout
# Poll every 200ms so we notice interrupts quickly
while proc.poll() is None:
if _interrupt_event.is_set():
# User sent a new message — kill the process tree and return
# what we have so far
try:
os.killpg(os.getpgid(proc.pid), signal.SIGTERM)
except (ProcessLookupError, PermissionError):
proc.kill()
reader.join(timeout=2)
output = "".join(_output_chunks)
return {
"output": output + "\n[Command interrupted — user sent a new message]",
"returncode": 130 # Standard interrupted exit code
}
if time.monotonic() > deadline:
# Timeout — kill process tree
try:
os.killpg(os.getpgid(proc.pid), signal.SIGTERM)
except (ProcessLookupError, PermissionError):
proc.kill()
reader.join(timeout=2)
return {"output": f"Command timed out after {effective_timeout}s", "returncode": 124}
# Short sleep to avoid busy-waiting
time.sleep(0.2)
# Process finished — wait for reader to drain remaining output
reader.join(timeout=5)
return {"output": "".join(_output_chunks), "returncode": proc.returncode}
except Exception as e:
return {"output": f"Execution error: {str(e)}", "returncode": 1}
def cleanup(self):
"""No cleanup needed for local environment."""
pass
def stop(self):
"""Alias for cleanup."""
pass
class _SingularityEnvironment:
"""
Persistent Singularity/Apptainer container environment.
Uses `apptainer instance` to create a long-running container that persists
state (files, installs, env changes) across all commands within a task.
The model experiences this as a real Linux VM.
Features:
- Persistent filesystem: files created in one command are visible in the next
- Package installs persist: pip/apt installs survive across tool calls
- Full isolation: --containall gives PID, IPC, and environment isolation
- Writable tmpfs overlay: full root filesystem is writable (RAM-backed)
- Automatic SIF caching: docker:// images converted to SIF once, reused forever
"""
def __init__(self, image: str, cwd: str = "/root", timeout: int = 60):
self.cwd = cwd
self.timeout = timeout
# Use apptainer if available, otherwise singularity
self.executable = "apptainer" if shutil.which("apptainer") else "singularity"
# Get or build SIF from docker:// URL (fast if already cached)
self.image = _get_or_build_sif(image, self.executable)
# Create unique instance name (must be alphanumeric + underscores)
self.instance_id = f"hermes_{uuid.uuid4().hex[:12]}"
self._instance_started = False
# Start the persistent instance
self._start_instance()
def _start_instance(self):
"""Start a persistent apptainer instance.
The instance runs as a background process. All subsequent execute() calls
run commands inside this same instance, so state persists across calls.
"""
cmd = [
self.executable, "instance", "start",
"--writable-tmpfs", # RAM-backed writable overlay on read-only SIF
"--containall", # Full isolation: PID, IPC, environment, filesystem
str(self.image),
self.instance_id,
]
try:
result = subprocess.run(
cmd,
capture_output=True,
text=True,
timeout=120, # 2 min for instance startup
)
if result.returncode != 0:
raise RuntimeError(f"Failed to start instance: {result.stderr}")
self._instance_started = True
logger.info("Singularity instance %s started (persistent container)", self.instance_id)
except subprocess.TimeoutExpired:
raise RuntimeError("Instance start timed out")
def execute(self, command: str, cwd: str = "", *, timeout: int | None = None,
stdin_data: str | None = None) -> dict:
"""Execute a command in the persistent Singularity instance.
All commands run in the same container, so files, installs, and
environment changes persist between calls.
"""
if not self._instance_started:
return {"output": "Instance not started", "returncode": -1}
cmd = [self.executable, "exec"]
# Set working directory
work_dir = cwd or self.cwd
cmd.extend(["--pwd", work_dir])
# Connect to the running instance
cmd.append(f"instance://{self.instance_id}")
# Transform sudo commands if SUDO_PASSWORD is available
exec_command = _transform_sudo_command(command)
# Execute the command
cmd.extend(["bash", "-c", exec_command])
run_kwargs = {
"text": True,
"timeout": timeout or self.timeout,
"encoding": "utf-8",
"errors": "replace",
"stdout": subprocess.PIPE,
"stderr": subprocess.STDOUT,
}
if stdin_data is not None:
run_kwargs["input"] = stdin_data
else:
run_kwargs["stdin"] = subprocess.DEVNULL
try:
result = subprocess.run(cmd, **run_kwargs)
return {"output": result.stdout, "returncode": result.returncode}
except subprocess.TimeoutExpired:
return {"output": f"Command timed out after {timeout or self.timeout}s", "returncode": 124}
def cleanup(self):
"""Stop the persistent instance and clean up."""
if self._instance_started:
try:
subprocess.run(
[self.executable, "instance", "stop", self.instance_id],
capture_output=True,
text=True,
timeout=30,
)
logger.info("Singularity instance %s stopped", self.instance_id)
except Exception as e:
logger.warning("Failed to stop Singularity instance %s: %s", self.instance_id, e)
self._instance_started = False
def stop(self):
"""Alias for cleanup."""
self.cleanup()
def __del__(self):
"""Cleanup on destruction."""
try:
self.cleanup()
except Exception:
pass
class _SSHEnvironment:
"""
SSH-based remote execution environment.
Runs commands on a remote machine over SSH, keeping the agent code
completely isolated from the execution environment. Uses SSH ControlMaster
for connection persistence (faster subsequent commands).
Security benefits:
- Agent cannot modify its own code
- Remote machine acts as a sandbox
- Clear separation between agent and execution environment
"""
def __init__(self, host: str, user: str, cwd: str = "/tmp", timeout: int = 60,
port: int = 22, key_path: str = ""):
self.host = host
self.user = user
self.cwd = cwd
self.timeout = timeout
self.port = port
self.key_path = key_path
# Create control socket directory for connection persistence
self.control_dir = Path(tempfile.gettempdir()) / "hermes-ssh"
self.control_dir.mkdir(parents=True, exist_ok=True)
self.control_socket = self.control_dir / f"{user}@{host}:{port}.sock"
# Test connection and establish ControlMaster
self._establish_connection()
def _build_ssh_command(self, extra_args: list = None) -> list:
"""Build base SSH command with connection options."""
cmd = ["ssh"]
# Connection multiplexing for performance
cmd.extend(["-o", f"ControlPath={self.control_socket}"])
cmd.extend(["-o", "ControlMaster=auto"])
cmd.extend(["-o", "ControlPersist=300"]) # Keep connection alive for 5 min
# Standard options
cmd.extend(["-o", "BatchMode=yes"]) # No password prompts
cmd.extend(["-o", "StrictHostKeyChecking=accept-new"]) # Accept new hosts
cmd.extend(["-o", "ConnectTimeout=10"])
# Port
if self.port != 22:
cmd.extend(["-p", str(self.port)])
# Private key
if self.key_path:
cmd.extend(["-i", self.key_path])
# Extra args (like -t for TTY)
if extra_args:
cmd.extend(extra_args)
# Target
cmd.append(f"{self.user}@{self.host}")
return cmd
def _establish_connection(self):
"""Test SSH connection and establish ControlMaster."""
cmd = self._build_ssh_command()
cmd.append("echo 'SSH connection established'")
try:
result = subprocess.run(
cmd,
capture_output=True,
text=True,
timeout=15
)
if result.returncode != 0:
error_msg = result.stderr.strip() or result.stdout.strip()
raise RuntimeError(f"SSH connection failed: {error_msg}")
except subprocess.TimeoutExpired:
raise RuntimeError(f"SSH connection to {self.user}@{self.host} timed out")
def execute(self, command: str, cwd: str = "", *, timeout: int | None = None,
stdin_data: str | None = None) -> dict:
"""Execute a command on the remote host via SSH."""
work_dir = cwd or self.cwd
effective_timeout = timeout or self.timeout
# Transform sudo commands if SUDO_PASSWORD is available
exec_command = _transform_sudo_command(command)
# Wrap command to run in the correct directory
wrapped_command = f'cd {work_dir} && {exec_command}'
cmd = self._build_ssh_command()
cmd.extend(["bash", "-c", wrapped_command])
run_kwargs = {
"text": True,
"timeout": effective_timeout,
"encoding": "utf-8",
"errors": "replace",
"stdout": subprocess.PIPE,
"stderr": subprocess.STDOUT,
}
if stdin_data is not None:
run_kwargs["input"] = stdin_data
else:
run_kwargs["stdin"] = subprocess.DEVNULL
try:
result = subprocess.run(cmd, **run_kwargs)
return {"output": result.stdout, "returncode": result.returncode}
except subprocess.TimeoutExpired:
return {"output": f"Command timed out after {effective_timeout}s", "returncode": 124}
except Exception as e:
return {"output": f"SSH execution error: {str(e)}", "returncode": 1}
def cleanup(self):
"""Close the SSH ControlMaster connection."""
if self.control_socket.exists():
try:
# Send exit command to ControlMaster
cmd = ["ssh", "-o", f"ControlPath={self.control_socket}", "-O", "exit",
f"{self.user}@{self.host}"]
subprocess.run(cmd, capture_output=True, timeout=5)
except (OSError, subprocess.SubprocessError):
pass
# Remove socket file
try:
self.control_socket.unlink()
except OSError:
pass
def stop(self):
"""Alias for cleanup."""
self.cleanup()
def __del__(self):
"""Cleanup on destruction."""
try:
self.cleanup()
except Exception:
pass
class _DockerEnvironment:
"""
Docker execution environment wrapper with sudo support and non-blocking stdin.
Wraps mini-swe-agent's DockerEnvironment but adds:
- stdin=DEVNULL to prevent hanging on interactive prompts
- SUDO_PASSWORD support via _transform_sudo_command
"""
def __init__(self, image: str, cwd: str = "/", timeout: int = 60):
from minisweagent.environments.docker import DockerEnvironment
self._inner = DockerEnvironment(image=image, cwd=cwd, timeout=timeout)
self.cwd = cwd
self.timeout = timeout
def execute(self, command: str, cwd: str = "", *, timeout: int | None = None,
stdin_data: str | None = None) -> dict:
"""Execute a command in the Docker container with sudo support."""
# Transform sudo commands if SUDO_PASSWORD is available
exec_command = _transform_sudo_command(command)
work_dir = cwd or self.cwd
effective_timeout = timeout or self.timeout
# Get container_id from inner environment
assert self._inner.container_id, "Container not started"
cmd = [self._inner.config.executable, "exec"]
if stdin_data is not None:
cmd.append("-i") # Enable stdin piping into the container
cmd.extend(["-w", work_dir])
for key in self._inner.config.forward_env:
if (value := os.getenv(key)) is not None:
cmd.extend(["-e", f"{key}={value}"])
for key, value in self._inner.config.env.items():
cmd.extend(["-e", f"{key}={value}"])
cmd.extend([self._inner.container_id, "bash", "-lc", exec_command])
run_kwargs = {
"text": True,
"timeout": effective_timeout,
"encoding": "utf-8",
"errors": "replace",
"stdout": subprocess.PIPE,
"stderr": subprocess.STDOUT,
}
if stdin_data is not None:
run_kwargs["input"] = stdin_data
else:
run_kwargs["stdin"] = subprocess.DEVNULL
try:
result = subprocess.run(cmd, **run_kwargs)
return {"output": result.stdout, "returncode": result.returncode}
except subprocess.TimeoutExpired:
return {"output": f"Command timed out after {effective_timeout}s", "returncode": 124}
def cleanup(self):
"""Cleanup the Docker container."""
self._inner.cleanup()
def stop(self):
"""Alias for cleanup."""
self.cleanup()
def __del__(self):
"""Cleanup on destruction."""
try:
self.cleanup()
except Exception:
pass
class _ModalEnvironment:
"""
Modal cloud execution environment wrapper with sudo support.
Wraps mini-swe-agent's SwerexModalEnvironment but adds:
- SUDO_PASSWORD support via _transform_sudo_command
- Automatic async-safety patches (applied once, before first use)
The patches replace SwerexModalEnvironment's asyncio.run() calls with a
background thread approach, making it safe to use inside any event loop
(e.g., Atropos). Applied here at the point of use rather than relying on
import-time side effects, so ALL callers get the fix automatically.
"""
# Class-level flag: patches only need to be applied once
_patches_applied = False
def __init__(self, image: str, cwd: str = "/root", timeout: int = 60):
# Ensure async-safety patches are applied before creating any
# SwerexModalEnvironment instance. This is the single authoritative
# place -- no other module needs to call apply_patches() for Modal.
if not _ModalEnvironment._patches_applied:
try:
from environments.patches import apply_patches
apply_patches()
except ImportError:
pass # patches module not available (standalone use)
_ModalEnvironment._patches_applied = True
from minisweagent.environments.extra.swerex_modal import SwerexModalEnvironment
# Generous startup timeout: sandbox creation can take 30-60s for cold images,
# and the SWE-ReX runtime needs another 10-30s to boot inside it.
self._inner = SwerexModalEnvironment(
image=image, cwd=cwd, timeout=timeout,
startup_timeout=180.0,
runtime_timeout=3600.0,
)
self.cwd = cwd
self.timeout = timeout
def execute(self, command: str, cwd: str = "", *, timeout: int | None = None,
stdin_data: str | None = None) -> dict:
"""Execute a command in Modal with sudo support.
Modal uses HTTP transport (no execve), so there's no ARG_MAX limit.
When stdin_data is provided, we embed it as a heredoc since there's
no process-level stdin pipe to the cloud sandbox.
"""
if stdin_data is not None:
marker = f"HERMES_EOF_{uuid.uuid4().hex[:8]}"
while marker in stdin_data:
marker = f"HERMES_EOF_{uuid.uuid4().hex[:8]}"
command = f"{command} << '{marker}'\n{stdin_data}\n{marker}"
# Transform sudo commands if SUDO_PASSWORD is available
exec_command = _transform_sudo_command(command)
# Delegate to inner environment with transformed command
return self._inner.execute(exec_command, cwd=cwd, timeout=timeout)
def cleanup(self):
"""Cleanup the Modal deployment."""
if hasattr(self._inner, 'stop'):
self._inner.stop()
def stop(self):
"""Stop the Modal deployment."""
self.cleanup()
def __del__(self):
"""Cleanup on destruction."""
try:
self.cleanup()
except Exception:
pass
# Environment classes now live in tools/environments/
from tools.environments.local import LocalEnvironment as _LocalEnvironment
from tools.environments.singularity import SingularityEnvironment as _SingularityEnvironment
from tools.environments.ssh import SSHEnvironment as _SSHEnvironment
from tools.environments.docker import DockerEnvironment as _DockerEnvironment
from tools.environments.modal import ModalEnvironment as _ModalEnvironment
# Tool description for LLM