Files
deer-flow/backend/packages/harness/deerflow/sandbox/tools.py
DanielWalnut d119214fee feat(harness): integration ACP agent tool (#1344)
* refactor: extract shared utils to break harness→app cross-layer imports

Move _validate_skill_frontmatter to src/skills/validation.py and
CONVERTIBLE_EXTENSIONS + convert_file_to_markdown to src/utils/file_conversion.py.
This eliminates the two reverse dependencies from client.py (harness layer)
into gateway/routers/ (app layer), preparing for the harness/app package split.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor: split backend/src into harness (deerflow.*) and app (app.*)

Physically split the monolithic backend/src/ package into two layers:

- **Harness** (`packages/harness/deerflow/`): publishable agent framework
  package with import prefix `deerflow.*`. Contains agents, sandbox, tools,
  models, MCP, skills, config, and all core infrastructure.

- **App** (`app/`): unpublished application code with import prefix `app.*`.
  Contains gateway (FastAPI REST API) and channels (IM integrations).

Key changes:
- Move 13 harness modules to packages/harness/deerflow/ via git mv
- Move gateway + channels to app/ via git mv
- Rename all imports: src.* → deerflow.* (harness) / app.* (app layer)
- Set up uv workspace with deerflow-harness as workspace member
- Update langgraph.json, config.example.yaml, all scripts, Docker files
- Add build-system (hatchling) to harness pyproject.toml
- Add PYTHONPATH=. to gateway startup commands for app.* resolution
- Update ruff.toml with known-first-party for import sorting
- Update all documentation to reflect new directory structure

Boundary rule enforced: harness code never imports from app.
All 429 tests pass. Lint clean.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* chore: add harness→app boundary check test and update docs

Add test_harness_boundary.py that scans all Python files in
packages/harness/deerflow/ and fails if any `from app.*` or
`import app.*` statement is found. This enforces the architectural
rule that the harness layer never depends on the app layer.

Update CLAUDE.md to document the harness/app split architecture,
import conventions, and the boundary enforcement test.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* feat: add config versioning with auto-upgrade on startup

When config.example.yaml schema changes, developers' local config.yaml
files can silently become outdated. This adds a config_version field and
auto-upgrade mechanism so breaking changes (like src.* → deerflow.*
renames) are applied automatically before services start.

- Add config_version: 1 to config.example.yaml
- Add startup version check warning in AppConfig.from_file()
- Add scripts/config-upgrade.sh with migration registry for value replacements
- Add `make config-upgrade` target
- Auto-run config-upgrade in serve.sh and start-daemon.sh before starting services
- Add config error hints in service failure messages

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix comments

* fix: update src.* import in test_sandbox_tools_security to deerflow.*

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: handle empty config and search parent dirs for config.example.yaml

Address Copilot review comments on PR #1131:
- Guard against yaml.safe_load() returning None for empty config files
- Search parent directories for config.example.yaml instead of only
  looking next to config.yaml, fixing detection in common setups

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: correct skills root path depth and config_version type coercion

- loader.py: fix get_skills_root_path() to use 5 parent levels (was 3)
  after harness split, file lives at packages/harness/deerflow/skills/
  so parent×3 resolved to backend/packages/harness/ instead of backend/
- app_config.py: coerce config_version to int() before comparison in
  _check_config_version() to prevent TypeError when YAML stores value
  as string (e.g. config_version: "1")
- tests: add regression tests for both fixes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix: update test imports from src.* to deerflow.*/app.* after harness refactor

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat(harness): add tool-first ACP agent invocation (#37)

* feat(harness): add tool-first ACP agent invocation

* build(harness): make ACP dependency required

* fix(harness): address ACP review feedback

* feat(harness): decouple ACP agent workspace from thread data

ACP agents (codex, claude-code) previously used per-thread workspace
directories, causing path resolution complexity and coupling task
execution to DeerFlow's internal thread data layout. This change:

- Replace _resolve_cwd() with a fixed _get_work_dir() that always uses
  {base_dir}/acp-workspace/, eliminating virtual path translation and
  thread_id lookups
- Introduce /mnt/acp-workspace virtual path for lead agent read-only
  access to ACP agent output files (same pattern as /mnt/skills)
- Add security guards: read-only validation, path traversal prevention,
  command path allowlisting, and output masking for acp-workspace
- Update system prompt and tool description to guide LLM: send
  self-contained tasks to ACP agents, copy results via /mnt/acp-workspace
- Add 11 new security tests for ACP workspace path handling

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* refactor(prompt): inject ACP section only when ACP agents are configured

The ACP agent guidance in the system prompt is now conditionally built
by _build_acp_section(), which checks get_acp_agents() and returns an
empty string when no ACP agents are configured. This avoids polluting
the prompt with irrelevant instructions for users who don't use ACP.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix lint

* fix(harness): address Copilot review comments on sandbox path handling and ACP tool

- local_sandbox: fix path-segment boundary bug in _resolve_path (== or startswith +"/")
  and add lookahead in _resolve_paths_in_command regex to prevent /mnt/skills matching
  inside /mnt/skills-extra
- local_sandbox_provider: replace print() with logger.warning(..., exc_info=True)
- invoke_acp_agent_tool: guard getattr(option, "optionId") with None default + continue;
  move full prompt from INFO to DEBUG level (truncated to 200 chars)
- sandbox/tools: fix _get_acp_workspace_host_path docstring to match implementation;
  remove misleading "read-only" language from validate_local_bash_command_paths

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(acp): thread-isolated workspaces, permission guardrail, and ContextVar registry

P1.1 – ACP workspace thread isolation
- Add `Paths.acp_workspace_dir(thread_id)` for per-thread paths
- `_get_work_dir(thread_id)` in invoke_acp_agent_tool now uses
  `{base_dir}/threads/{thread_id}/acp-workspace/`; falls back to
  global workspace when thread_id is absent or invalid
- `_invoke` extracts thread_id from `RunnableConfig` via
  `Annotated[RunnableConfig, InjectedToolArg]`
- `sandbox/tools.py`: `_get_acp_workspace_host_path(thread_id)`,
  `_resolve_acp_workspace_path(path, thread_id)`, and all callers
  (`replace_virtual_paths_in_command`, `mask_local_paths_in_output`,
  `ls_tool`, `read_file_tool`) now resolve ACP paths per-thread

P1.2 – ACP permission guardrail
- New `auto_approve_permissions: bool = False` field in `ACPAgentConfig`
- `_build_permission_response(options, *, auto_approve: bool)` now
  defaults to deny; only approves when `auto_approve=True`
- Document field in `config.example.yaml`

P2 – Deferred tool registry race condition
- Replace module-level `_registry` global with `contextvars.ContextVar`
- Each asyncio request context gets its own registry; worker threads
  inherit the context automatically via `loop.run_in_executor`
- Expose `get_deferred_registry` / `set_deferred_registry` /
  `reset_deferred_registry` helpers

Tests: 831 pass (57 for affected modules, 3 new tests)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(sandbox): mount /mnt/acp-workspace in docker sandbox container

The AioSandboxProvider was not mounting the ACP workspace into the
sandbox container, so /mnt/acp-workspace was inaccessible when the lead
agent tried to read ACP results in docker mode.

Changes:
- `ensure_thread_dirs`: also create `acp-workspace/` (chmod 0o777) so
  the directory exists before the sandbox container starts — required
  for Docker volume mounts
- `_get_thread_mounts`: add read-only `/mnt/acp-workspace` mount using
  the per-thread host path (`host_paths.acp_workspace_dir(thread_id)`)
- Update stale CLAUDE.md description (was "fixed global workspace")

Tests: `test_aio_sandbox_provider.py` (4 new tests)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(lint): remove unused imports in test_aio_sandbox_provider

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix config

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-26 14:20:18 +08:00

880 lines
34 KiB
Python

import re
from pathlib import Path
from langchain.tools import ToolRuntime, tool
from langgraph.typing import ContextT
from deerflow.agents.thread_state import ThreadDataState, ThreadState
from deerflow.config.paths import VIRTUAL_PATH_PREFIX
from deerflow.sandbox.exceptions import (
SandboxError,
SandboxNotFoundError,
SandboxRuntimeError,
)
from deerflow.sandbox.sandbox import Sandbox
from deerflow.sandbox.sandbox_provider import get_sandbox_provider
_ABSOLUTE_PATH_PATTERN = re.compile(r"(?<![:\w])/(?:[^\s\"'`;&|<>()]+)")
_LOCAL_BASH_SYSTEM_PATH_PREFIXES = (
"/bin/",
"/usr/bin/",
"/usr/sbin/",
"/sbin/",
"/opt/homebrew/bin/",
"/dev/",
)
_DEFAULT_SKILLS_CONTAINER_PATH = "/mnt/skills"
_ACP_WORKSPACE_VIRTUAL_PATH = "/mnt/acp-workspace"
def _get_skills_container_path() -> str:
"""Get the skills container path from config, with fallback to default.
Result is cached after the first successful config load. If config loading
fails the default is returned *without* caching so that a later call can
pick up the real value once the config is available.
"""
cached = getattr(_get_skills_container_path, "_cached", None)
if cached is not None:
return cached
try:
from deerflow.config import get_app_config
value = get_app_config().skills.container_path
_get_skills_container_path._cached = value # type: ignore[attr-defined]
return value
except Exception:
return _DEFAULT_SKILLS_CONTAINER_PATH
def _get_skills_host_path() -> str | None:
"""Get the skills host filesystem path from config.
Returns None if the skills directory does not exist or config cannot be
loaded. Only successful lookups are cached; failures are retried on the
next call so that a transiently unavailable skills directory does not
permanently disable skills access.
"""
cached = getattr(_get_skills_host_path, "_cached", None)
if cached is not None:
return cached
try:
from deerflow.config import get_app_config
config = get_app_config()
skills_path = config.skills.get_skills_path()
if skills_path.exists():
value = str(skills_path)
_get_skills_host_path._cached = value # type: ignore[attr-defined]
return value
except Exception:
pass
return None
def _is_skills_path(path: str) -> bool:
"""Check if a path is under the skills container path."""
skills_prefix = _get_skills_container_path()
return path == skills_prefix or path.startswith(f"{skills_prefix}/")
def _resolve_skills_path(path: str) -> str:
"""Resolve a virtual skills path to a host filesystem path.
Args:
path: Virtual skills path (e.g. /mnt/skills/public/bootstrap/SKILL.md)
Returns:
Resolved host path.
Raises:
FileNotFoundError: If skills directory is not configured or doesn't exist.
"""
skills_container = _get_skills_container_path()
skills_host = _get_skills_host_path()
if skills_host is None:
raise FileNotFoundError(f"Skills directory not available for path: {path}")
if path == skills_container:
return skills_host
relative = path[len(skills_container) :].lstrip("/")
return str(Path(skills_host) / relative) if relative else skills_host
def _is_acp_workspace_path(path: str) -> bool:
"""Check if a path is under the ACP workspace virtual path."""
return path == _ACP_WORKSPACE_VIRTUAL_PATH or path.startswith(f"{_ACP_WORKSPACE_VIRTUAL_PATH}/")
def _extract_thread_id_from_thread_data(thread_data: "ThreadDataState | None") -> str | None:
"""Extract thread_id from thread_data by inspecting workspace_path.
The workspace_path has the form
``{base_dir}/threads/{thread_id}/user-data/workspace``, so
``Path(workspace_path).parent.parent.name`` yields the thread_id.
"""
if thread_data is None:
return None
workspace_path = thread_data.get("workspace_path")
if not workspace_path:
return None
try:
# {base_dir}/threads/{thread_id}/user-data/workspace → parent.parent = threads/{thread_id}
return Path(workspace_path).parent.parent.name
except Exception:
return None
def _get_acp_workspace_host_path(thread_id: str | None = None) -> str | None:
"""Get the ACP workspace host filesystem path.
When *thread_id* is provided, returns the per-thread workspace
``{base_dir}/threads/{thread_id}/acp-workspace/`` (not cached — the
directory is created on demand by ``invoke_acp_agent_tool``).
Falls back to the global ``{base_dir}/acp-workspace/`` when *thread_id*
is ``None``; that result is cached after the first successful resolution.
Returns ``None`` if the directory does not exist.
"""
if thread_id is not None:
try:
from deerflow.config.paths import get_paths
host_path = get_paths().acp_workspace_dir(thread_id)
if host_path.exists():
return str(host_path)
except Exception:
pass
return None
cached = getattr(_get_acp_workspace_host_path, "_cached", None)
if cached is not None:
return cached
try:
from deerflow.config.paths import get_paths
host_path = get_paths().base_dir / "acp-workspace"
if host_path.exists():
value = str(host_path)
_get_acp_workspace_host_path._cached = value # type: ignore[attr-defined]
return value
except Exception:
pass
return None
def _resolve_acp_workspace_path(path: str, thread_id: str | None = None) -> str:
"""Resolve a virtual ACP workspace path to a host filesystem path.
Args:
path: Virtual path (e.g. /mnt/acp-workspace/hello_world.py)
thread_id: Current thread ID for per-thread workspace resolution.
When ``None``, falls back to the global workspace.
Returns:
Resolved host path.
Raises:
FileNotFoundError: If ACP workspace directory does not exist.
PermissionError: If path traversal is detected.
"""
_reject_path_traversal(path)
host_path = _get_acp_workspace_host_path(thread_id)
if host_path is None:
raise FileNotFoundError(f"ACP workspace directory not available for path: {path}")
if path == _ACP_WORKSPACE_VIRTUAL_PATH:
return host_path
relative = path[len(_ACP_WORKSPACE_VIRTUAL_PATH) :].lstrip("/")
if not relative:
return host_path
resolved = Path(host_path).resolve() / relative
# Ensure resolved path stays inside the ACP workspace
try:
resolved.resolve().relative_to(Path(host_path).resolve())
except ValueError:
raise PermissionError("Access denied: path traversal detected")
return str(resolved)
def _path_variants(path: str) -> set[str]:
return {path, path.replace("\\", "/"), path.replace("/", "\\")}
def _sanitize_error(error: Exception, runtime: "ToolRuntime[ContextT, ThreadState] | None" = None) -> str:
"""Sanitize an error message to avoid leaking host filesystem paths.
In local-sandbox mode, resolved host paths in the error string are masked
back to their virtual equivalents so that user-visible output never exposes
the host directory layout.
"""
msg = f"{type(error).__name__}: {error}"
if runtime is not None and is_local_sandbox(runtime):
thread_data = get_thread_data(runtime)
msg = mask_local_paths_in_output(msg, thread_data)
return msg
def replace_virtual_path(path: str, thread_data: ThreadDataState | None) -> str:
"""Replace virtual /mnt/user-data paths with actual thread data paths.
Mapping:
/mnt/user-data/workspace/* -> thread_data['workspace_path']/*
/mnt/user-data/uploads/* -> thread_data['uploads_path']/*
/mnt/user-data/outputs/* -> thread_data['outputs_path']/*
Args:
path: The path that may contain virtual path prefix.
thread_data: The thread data containing actual paths.
Returns:
The path with virtual prefix replaced by actual path.
"""
if thread_data is None:
return path
mappings = _thread_virtual_to_actual_mappings(thread_data)
if not mappings:
return path
# Longest-prefix-first replacement with segment-boundary checks.
for virtual_base, actual_base in sorted(mappings.items(), key=lambda item: len(item[0]), reverse=True):
if path == virtual_base:
return actual_base
if path.startswith(f"{virtual_base}/"):
rest = path[len(virtual_base) :].lstrip("/")
return str(Path(actual_base) / rest) if rest else actual_base
return path
def _thread_virtual_to_actual_mappings(thread_data: ThreadDataState) -> dict[str, str]:
"""Build virtual-to-actual path mappings for a thread."""
mappings: dict[str, str] = {}
workspace = thread_data.get("workspace_path")
uploads = thread_data.get("uploads_path")
outputs = thread_data.get("outputs_path")
if workspace:
mappings[f"{VIRTUAL_PATH_PREFIX}/workspace"] = workspace
if uploads:
mappings[f"{VIRTUAL_PATH_PREFIX}/uploads"] = uploads
if outputs:
mappings[f"{VIRTUAL_PATH_PREFIX}/outputs"] = outputs
# Also map the virtual root when all known dirs share the same parent.
actual_dirs = [Path(p) for p in (workspace, uploads, outputs) if p]
if actual_dirs:
common_parent = str(Path(actual_dirs[0]).parent)
if all(str(path.parent) == common_parent for path in actual_dirs):
mappings[VIRTUAL_PATH_PREFIX] = common_parent
return mappings
def _thread_actual_to_virtual_mappings(thread_data: ThreadDataState) -> dict[str, str]:
"""Build actual-to-virtual mappings for output masking."""
return {actual: virtual for virtual, actual in _thread_virtual_to_actual_mappings(thread_data).items()}
def mask_local_paths_in_output(output: str, thread_data: ThreadDataState | None) -> str:
"""Mask host absolute paths from local sandbox output using virtual paths.
Handles user-data paths (per-thread), skills paths, and ACP workspace paths (global).
"""
result = output
# Mask skills host paths
skills_host = _get_skills_host_path()
skills_container = _get_skills_container_path()
if skills_host:
raw_base = str(Path(skills_host))
resolved_base = str(Path(skills_host).resolve())
for base in _path_variants(raw_base) | _path_variants(resolved_base):
escaped = re.escape(base).replace(r"\\", r"[/\\]")
pattern = re.compile(escaped + r"(?:[/\\][^\s\"';&|<>()]*)?")
def replace_skills(match: re.Match, _base: str = base) -> str:
matched_path = match.group(0)
if matched_path == _base:
return skills_container
relative = matched_path[len(_base) :].lstrip("/\\")
return f"{skills_container}/{relative}" if relative else skills_container
result = pattern.sub(replace_skills, result)
# Mask ACP workspace host paths
_thread_id = _extract_thread_id_from_thread_data(thread_data)
acp_host = _get_acp_workspace_host_path(_thread_id)
if acp_host:
raw_base = str(Path(acp_host))
resolved_base = str(Path(acp_host).resolve())
for base in _path_variants(raw_base) | _path_variants(resolved_base):
escaped = re.escape(base).replace(r"\\", r"[/\\]")
pattern = re.compile(escaped + r"(?:[/\\][^\s\"';&|<>()]*)?")
def replace_acp(match: re.Match, _base: str = base) -> str:
matched_path = match.group(0)
if matched_path == _base:
return _ACP_WORKSPACE_VIRTUAL_PATH
relative = matched_path[len(_base) :].lstrip("/\\")
return f"{_ACP_WORKSPACE_VIRTUAL_PATH}/{relative}" if relative else _ACP_WORKSPACE_VIRTUAL_PATH
result = pattern.sub(replace_acp, result)
# Mask user-data host paths
if thread_data is None:
return result
mappings = _thread_actual_to_virtual_mappings(thread_data)
if not mappings:
return result
for actual_base, virtual_base in sorted(mappings.items(), key=lambda item: len(item[0]), reverse=True):
raw_base = str(Path(actual_base))
resolved_base = str(Path(actual_base).resolve())
for base in _path_variants(raw_base) | _path_variants(resolved_base):
escaped_actual = re.escape(base).replace(r"\\", r"[/\\]")
pattern = re.compile(escaped_actual + r"(?:[/\\][^\s\"';&|<>()]*)?")
def replace_match(match: re.Match, _base: str = base, _virtual: str = virtual_base) -> str:
matched_path = match.group(0)
if matched_path == _base:
return _virtual
relative = matched_path[len(_base) :].lstrip("/\\")
return f"{_virtual}/{relative}" if relative else _virtual
result = pattern.sub(replace_match, result)
return result
def _reject_path_traversal(path: str) -> None:
"""Reject paths that contain '..' segments to prevent directory traversal."""
# Normalise to forward slashes, then check for '..' segments.
normalised = path.replace("\\", "/")
for segment in normalised.split("/"):
if segment == "..":
raise PermissionError("Access denied: path traversal detected")
def validate_local_tool_path(path: str, thread_data: ThreadDataState | None, *, read_only: bool = False) -> None:
"""Validate that a virtual path is allowed for local-sandbox access.
This function is a security gate — it checks whether *path* may be
accessed and raises on violation. It does **not** resolve the virtual
path to a host path; callers are responsible for resolution via
``_resolve_and_validate_user_data_path`` or ``_resolve_skills_path``.
Allowed virtual-path families:
- ``/mnt/user-data/*`` — always allowed (read + write)
- ``/mnt/skills/*`` — allowed only when *read_only* is True
- ``/mnt/acp-workspace/*`` — allowed only when *read_only* is True
Args:
path: The virtual path to validate.
thread_data: Thread data (must be present for local sandbox).
read_only: When True, skills and ACP workspace paths are permitted.
Raises:
SandboxRuntimeError: If thread data is missing.
PermissionError: If the path is not allowed or contains traversal.
"""
if thread_data is None:
raise SandboxRuntimeError("Thread data not available for local sandbox")
_reject_path_traversal(path)
# Skills paths — read-only access only
if _is_skills_path(path):
if not read_only:
raise PermissionError(f"Write access to skills path is not allowed: {path}")
return
# ACP workspace paths — read-only access only
if _is_acp_workspace_path(path):
if not read_only:
raise PermissionError(f"Write access to ACP workspace is not allowed: {path}")
return
# User-data paths
if path.startswith(f"{VIRTUAL_PATH_PREFIX}/"):
return
raise PermissionError(f"Only paths under {VIRTUAL_PATH_PREFIX}/, {_get_skills_container_path()}/, or {_ACP_WORKSPACE_VIRTUAL_PATH}/ are allowed")
def _validate_resolved_user_data_path(resolved: Path, thread_data: ThreadDataState) -> None:
"""Verify that a resolved host path stays inside allowed per-thread roots.
Raises PermissionError if the path escapes workspace/uploads/outputs.
"""
allowed_roots = [
Path(p).resolve()
for p in (
thread_data.get("workspace_path"),
thread_data.get("uploads_path"),
thread_data.get("outputs_path"),
)
if p is not None
]
if not allowed_roots:
raise SandboxRuntimeError("No allowed local sandbox directories configured")
for root in allowed_roots:
try:
resolved.relative_to(root)
return
except ValueError:
continue
raise PermissionError("Access denied: path traversal detected")
def _resolve_and_validate_user_data_path(path: str, thread_data: ThreadDataState) -> str:
"""Resolve a /mnt/user-data virtual path and validate it stays in bounds.
Returns the resolved host path string.
"""
resolved_str = replace_virtual_path(path, thread_data)
resolved = Path(resolved_str).resolve()
_validate_resolved_user_data_path(resolved, thread_data)
return str(resolved)
def validate_local_bash_command_paths(command: str, thread_data: ThreadDataState | None) -> None:
"""Validate absolute paths in local-sandbox bash commands.
In local mode, commands must use virtual paths under /mnt/user-data for
user data access. Skills paths under /mnt/skills and ACP workspace paths
under /mnt/acp-workspace are allowed (path-traversal checks only; write
prevention for bash commands is not enforced here).
A small allowlist of common system path prefixes is kept for executable
and device references (e.g. /bin/sh, /dev/null).
"""
if thread_data is None:
raise SandboxRuntimeError("Thread data not available for local sandbox")
unsafe_paths: list[str] = []
for absolute_path in _ABSOLUTE_PATH_PATTERN.findall(command):
if absolute_path == VIRTUAL_PATH_PREFIX or absolute_path.startswith(f"{VIRTUAL_PATH_PREFIX}/"):
_reject_path_traversal(absolute_path)
continue
# Allow skills container path (resolved by tools.py before passing to sandbox)
if _is_skills_path(absolute_path):
_reject_path_traversal(absolute_path)
continue
# Allow ACP workspace path (path-traversal check only)
if _is_acp_workspace_path(absolute_path):
_reject_path_traversal(absolute_path)
continue
if any(absolute_path == prefix.rstrip("/") or absolute_path.startswith(prefix) for prefix in _LOCAL_BASH_SYSTEM_PATH_PREFIXES):
continue
unsafe_paths.append(absolute_path)
if unsafe_paths:
unsafe = ", ".join(sorted(dict.fromkeys(unsafe_paths)))
raise PermissionError(f"Unsafe absolute paths in command: {unsafe}. Use paths under {VIRTUAL_PATH_PREFIX}")
def replace_virtual_paths_in_command(command: str, thread_data: ThreadDataState | None) -> str:
"""Replace all virtual paths (/mnt/user-data, /mnt/skills, /mnt/acp-workspace) in a command string.
Args:
command: The command string that may contain virtual paths.
thread_data: The thread data containing actual paths.
Returns:
The command with all virtual paths replaced.
"""
result = command
# Replace skills paths
skills_container = _get_skills_container_path()
skills_host = _get_skills_host_path()
if skills_host and skills_container in result:
skills_pattern = re.compile(rf"{re.escape(skills_container)}(/[^\s\"';&|<>()]*)?")
def replace_skills_match(match: re.Match) -> str:
return _resolve_skills_path(match.group(0))
result = skills_pattern.sub(replace_skills_match, result)
# Replace ACP workspace paths
_thread_id = _extract_thread_id_from_thread_data(thread_data)
acp_host = _get_acp_workspace_host_path(_thread_id)
if acp_host and _ACP_WORKSPACE_VIRTUAL_PATH in result:
acp_pattern = re.compile(rf"{re.escape(_ACP_WORKSPACE_VIRTUAL_PATH)}(/[^\s\"';&|<>()]*)?")
def replace_acp_match(match: re.Match, _tid: str | None = _thread_id) -> str:
return _resolve_acp_workspace_path(match.group(0), _tid)
result = acp_pattern.sub(replace_acp_match, result)
# Replace user-data paths
if VIRTUAL_PATH_PREFIX in result and thread_data is not None:
pattern = re.compile(rf"{re.escape(VIRTUAL_PATH_PREFIX)}(/[^\s\"';&|<>()]*)?")
def replace_user_data_match(match: re.Match) -> str:
return replace_virtual_path(match.group(0), thread_data)
result = pattern.sub(replace_user_data_match, result)
return result
def get_thread_data(runtime: ToolRuntime[ContextT, ThreadState] | None) -> ThreadDataState | None:
"""Extract thread_data from runtime state."""
if runtime is None:
return None
if runtime.state is None:
return None
return runtime.state.get("thread_data")
def is_local_sandbox(runtime: ToolRuntime[ContextT, ThreadState] | None) -> bool:
"""Check if the current sandbox is a local sandbox.
Path replacement is only needed for local sandbox since aio sandbox
already has /mnt/user-data mounted in the container.
"""
if runtime is None:
return False
if runtime.state is None:
return False
sandbox_state = runtime.state.get("sandbox")
if sandbox_state is None:
return False
return sandbox_state.get("sandbox_id") == "local"
def sandbox_from_runtime(runtime: ToolRuntime[ContextT, ThreadState] | None = None) -> Sandbox:
"""Extract sandbox instance from tool runtime.
DEPRECATED: Use ensure_sandbox_initialized() for lazy initialization support.
This function assumes sandbox is already initialized and will raise error if not.
Raises:
SandboxRuntimeError: If runtime is not available or sandbox state is missing.
SandboxNotFoundError: If sandbox with the given ID cannot be found.
"""
if runtime is None:
raise SandboxRuntimeError("Tool runtime not available")
if runtime.state is None:
raise SandboxRuntimeError("Tool runtime state not available")
sandbox_state = runtime.state.get("sandbox")
if sandbox_state is None:
raise SandboxRuntimeError("Sandbox state not initialized in runtime")
sandbox_id = sandbox_state.get("sandbox_id")
if sandbox_id is None:
raise SandboxRuntimeError("Sandbox ID not found in state")
sandbox = get_sandbox_provider().get(sandbox_id)
if sandbox is None:
raise SandboxNotFoundError(f"Sandbox with ID '{sandbox_id}' not found", sandbox_id=sandbox_id)
runtime.context["sandbox_id"] = sandbox_id # Ensure sandbox_id is in context for downstream use
return sandbox
def ensure_sandbox_initialized(runtime: ToolRuntime[ContextT, ThreadState] | None = None) -> Sandbox:
"""Ensure sandbox is initialized, acquiring lazily if needed.
On first call, acquires a sandbox from the provider and stores it in runtime state.
Subsequent calls return the existing sandbox.
Thread-safety is guaranteed by the provider's internal locking mechanism.
Args:
runtime: Tool runtime containing state and context.
Returns:
Initialized sandbox instance.
Raises:
SandboxRuntimeError: If runtime is not available or thread_id is missing.
SandboxNotFoundError: If sandbox acquisition fails.
"""
if runtime is None:
raise SandboxRuntimeError("Tool runtime not available")
if runtime.state is None:
raise SandboxRuntimeError("Tool runtime state not available")
# Check if sandbox already exists in state
sandbox_state = runtime.state.get("sandbox")
if sandbox_state is not None:
sandbox_id = sandbox_state.get("sandbox_id")
if sandbox_id is not None:
sandbox = get_sandbox_provider().get(sandbox_id)
if sandbox is not None:
runtime.context["sandbox_id"] = sandbox_id # Ensure sandbox_id is in context for releasing in after_agent
return sandbox
# Sandbox was released, fall through to acquire new one
# Lazy acquisition: get thread_id and acquire sandbox
thread_id = runtime.context.get("thread_id") if runtime.context else None
if thread_id is None:
raise SandboxRuntimeError("Thread ID not available in runtime context")
provider = get_sandbox_provider()
sandbox_id = provider.acquire(thread_id)
# Update runtime state - this persists across tool calls
runtime.state["sandbox"] = {"sandbox_id": sandbox_id}
# Retrieve and return the sandbox
sandbox = provider.get(sandbox_id)
if sandbox is None:
raise SandboxNotFoundError("Sandbox not found after acquisition", sandbox_id=sandbox_id)
runtime.context["sandbox_id"] = sandbox_id # Ensure sandbox_id is in context for releasing in after_agent
return sandbox
def ensure_thread_directories_exist(runtime: ToolRuntime[ContextT, ThreadState] | None) -> None:
"""Ensure thread data directories (workspace, uploads, outputs) exist.
This function is called lazily when any sandbox tool is first used.
For local sandbox, it creates the directories on the filesystem.
For other sandboxes (like aio), directories are already mounted in the container.
Args:
runtime: Tool runtime containing state and context.
"""
if runtime is None:
return
# Only create directories for local sandbox
if not is_local_sandbox(runtime):
return
thread_data = get_thread_data(runtime)
if thread_data is None:
return
# Check if directories have already been created
if runtime.state.get("thread_directories_created"):
return
# Create the three directories
import os
for key in ["workspace_path", "uploads_path", "outputs_path"]:
path = thread_data.get(key)
if path:
os.makedirs(path, exist_ok=True)
# Mark as created to avoid redundant operations
runtime.state["thread_directories_created"] = True
@tool("bash", parse_docstring=True)
def bash_tool(runtime: ToolRuntime[ContextT, ThreadState], description: str, command: str) -> str:
"""Execute a bash command in a Linux environment.
- Use `python` to run Python code.
- Prefer a thread-local virtual environment in `/mnt/user-data/workspace/.venv`.
- Use `python -m pip` (inside the virtual environment) to install Python packages.
Args:
description: Explain why you are running this command in short words. ALWAYS PROVIDE THIS PARAMETER FIRST.
command: The bash command to execute. Always use absolute paths for files and directories.
"""
try:
sandbox = ensure_sandbox_initialized(runtime)
ensure_thread_directories_exist(runtime)
thread_data = get_thread_data(runtime)
if is_local_sandbox(runtime):
validate_local_bash_command_paths(command, thread_data)
command = replace_virtual_paths_in_command(command, thread_data)
output = sandbox.execute_command(command)
return mask_local_paths_in_output(output, thread_data)
return sandbox.execute_command(command)
except SandboxError as e:
return f"Error: {e}"
except PermissionError as e:
return f"Error: {e}"
except Exception as e:
return f"Error: Unexpected error executing command: {_sanitize_error(e, runtime)}"
@tool("ls", parse_docstring=True)
def ls_tool(runtime: ToolRuntime[ContextT, ThreadState], description: str, path: str) -> str:
"""List the contents of a directory up to 2 levels deep in tree format.
Args:
description: Explain why you are listing this directory in short words. ALWAYS PROVIDE THIS PARAMETER FIRST.
path: The **absolute** path to the directory to list.
"""
try:
sandbox = ensure_sandbox_initialized(runtime)
ensure_thread_directories_exist(runtime)
requested_path = path
if is_local_sandbox(runtime):
thread_data = get_thread_data(runtime)
validate_local_tool_path(path, thread_data, read_only=True)
if _is_skills_path(path):
path = _resolve_skills_path(path)
elif _is_acp_workspace_path(path):
path = _resolve_acp_workspace_path(path, _extract_thread_id_from_thread_data(thread_data))
else:
path = _resolve_and_validate_user_data_path(path, thread_data)
children = sandbox.list_dir(path)
if not children:
return "(empty)"
return "\n".join(children)
except SandboxError as e:
return f"Error: {e}"
except FileNotFoundError:
return f"Error: Directory not found: {requested_path}"
except PermissionError:
return f"Error: Permission denied: {requested_path}"
except Exception as e:
return f"Error: Unexpected error listing directory: {_sanitize_error(e, runtime)}"
@tool("read_file", parse_docstring=True)
def read_file_tool(
runtime: ToolRuntime[ContextT, ThreadState],
description: str,
path: str,
start_line: int | None = None,
end_line: int | None = None,
) -> str:
"""Read the contents of a text file. Use this to examine source code, configuration files, logs, or any text-based file.
Args:
description: Explain why you are reading this file in short words. ALWAYS PROVIDE THIS PARAMETER FIRST.
path: The **absolute** path to the file to read.
start_line: Optional starting line number (1-indexed, inclusive). Use with end_line to read a specific range.
end_line: Optional ending line number (1-indexed, inclusive). Use with start_line to read a specific range.
"""
try:
sandbox = ensure_sandbox_initialized(runtime)
ensure_thread_directories_exist(runtime)
requested_path = path
if is_local_sandbox(runtime):
thread_data = get_thread_data(runtime)
validate_local_tool_path(path, thread_data, read_only=True)
if _is_skills_path(path):
path = _resolve_skills_path(path)
elif _is_acp_workspace_path(path):
path = _resolve_acp_workspace_path(path, _extract_thread_id_from_thread_data(thread_data))
else:
path = _resolve_and_validate_user_data_path(path, thread_data)
content = sandbox.read_file(path)
if not content:
return "(empty)"
if start_line is not None and end_line is not None:
content = "\n".join(content.splitlines()[start_line - 1 : end_line])
return content
except SandboxError as e:
return f"Error: {e}"
except FileNotFoundError:
return f"Error: File not found: {requested_path}"
except PermissionError:
return f"Error: Permission denied reading file: {requested_path}"
except IsADirectoryError:
return f"Error: Path is a directory, not a file: {requested_path}"
except Exception as e:
return f"Error: Unexpected error reading file: {_sanitize_error(e, runtime)}"
@tool("write_file", parse_docstring=True)
def write_file_tool(
runtime: ToolRuntime[ContextT, ThreadState],
description: str,
path: str,
content: str,
append: bool = False,
) -> str:
"""Write text content to a file.
Args:
description: Explain why you are writing to this file in short words. ALWAYS PROVIDE THIS PARAMETER FIRST.
path: The **absolute** path to the file to write to. ALWAYS PROVIDE THIS PARAMETER SECOND.
content: The content to write to the file. ALWAYS PROVIDE THIS PARAMETER THIRD.
"""
try:
sandbox = ensure_sandbox_initialized(runtime)
ensure_thread_directories_exist(runtime)
requested_path = path
if is_local_sandbox(runtime):
thread_data = get_thread_data(runtime)
validate_local_tool_path(path, thread_data)
path = _resolve_and_validate_user_data_path(path, thread_data)
sandbox.write_file(path, content, append)
return "OK"
except SandboxError as e:
return f"Error: {e}"
except PermissionError:
return f"Error: Permission denied writing to file: {requested_path}"
except IsADirectoryError:
return f"Error: Path is a directory, not a file: {requested_path}"
except OSError as e:
return f"Error: Failed to write file '{requested_path}': {_sanitize_error(e, runtime)}"
except Exception as e:
return f"Error: Unexpected error writing file: {_sanitize_error(e, runtime)}"
@tool("str_replace", parse_docstring=True)
def str_replace_tool(
runtime: ToolRuntime[ContextT, ThreadState],
description: str,
path: str,
old_str: str,
new_str: str,
replace_all: bool = False,
) -> str:
"""Replace a substring in a file with another substring.
If `replace_all` is False (default), the substring to replace must appear **exactly once** in the file.
Args:
description: Explain why you are replacing the substring in short words. ALWAYS PROVIDE THIS PARAMETER FIRST.
path: The **absolute** path to the file to replace the substring in. ALWAYS PROVIDE THIS PARAMETER SECOND.
old_str: The substring to replace. ALWAYS PROVIDE THIS PARAMETER THIRD.
new_str: The new substring. ALWAYS PROVIDE THIS PARAMETER FOURTH.
replace_all: Whether to replace all occurrences of the substring. If False, only the first occurrence will be replaced. Default is False.
"""
try:
sandbox = ensure_sandbox_initialized(runtime)
ensure_thread_directories_exist(runtime)
requested_path = path
if is_local_sandbox(runtime):
thread_data = get_thread_data(runtime)
validate_local_tool_path(path, thread_data)
path = _resolve_and_validate_user_data_path(path, thread_data)
content = sandbox.read_file(path)
if not content:
return "OK"
if old_str not in content:
return f"Error: String to replace not found in file: {requested_path}"
if replace_all:
content = content.replace(old_str, new_str)
else:
content = content.replace(old_str, new_str, 1)
sandbox.write_file(path, content)
return "OK"
except SandboxError as e:
return f"Error: {e}"
except FileNotFoundError:
return f"Error: File not found: {requested_path}"
except PermissionError:
return f"Error: Permission denied accessing file: {requested_path}"
except Exception as e:
return f"Error: Unexpected error replacing string: {_sanitize_error(e, runtime)}"