Compare commits

...

1 Commits

Author SHA1 Message Date
Timmy (AI Agent)
f58f9b2368 feat(integration): Browser Use anti-detect + Graphify skill (#262)
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 1m7s
Browser Use provider enhancements:
- Anti-detect fingerprinting (BROWSER_USE_ANTI_DETECT=true, default on)
- CAPTCHA auto-solving (BROWSER_USE_CAPTCHA_SOLVING=true, default on)
- Persistent login/cookies (BROWSER_USE_PERSISTENT_LOGIN=true)
- Named profiles for cookie jar isolation (BROWSER_USE_PROFILE_NAME)
- Proxy country override (BROWSER_USE_PROXY_COUNTRY)
- Full feature reporting in session metadata

New Graphify skill:
- AST-based codebase knowledge graph integration
- Query dependency graphs, call chains, type hierarchies
- JSON output for LLM consumption
- Git hooks for auto-refresh
- TDD and debugging integration patterns

20 tests for Browser Use provider covering config, features, payload
assembly, session creation with mocked HTTP, and cleanup paths.

Closes #262
2026-04-13 18:18:39 -04:00
3 changed files with 526 additions and 15 deletions

View File

@@ -0,0 +1,182 @@
---
name: graphify
description: AST-based codebase knowledge graph for precise code understanding. Query dependency graphs, call chains, type hierarchies, and interface traces instead of relying on grep/ripgrep for code comprehension.
version: 1.0.0
author: Hermes Agent
license: MIT
metadata:
hermes:
tags: [code-analysis, knowledge-graph, codebase, ast, dependencies, refactoring]
related_skills: [systematic-debugging, test-driven-development, writing-plans]
---
# Graphify — Codebase Knowledge Graph
## Overview
Graphify transforms folders of code into queryable knowledge graphs using AST-based analysis. Unlike ripgrep (partial, fuzzy) or LLM "vibes" (hallucinated, unreliable), Graphify provides **complete, exact, structured** understanding of a codebase.
| Approach | Coverage | Precision | Structured |
|----------|----------|-----------|-----------|
| ripgrep | Partial | Fuzzy | No |
| LLM "vibes" | Hallucinated | Unreliable | No |
| **Graphify** | Complete | Exact | Yes |
## Supported Languages
Python, TypeScript, JavaScript, Go, Java, Kotlin, Rust, C++
## Installation
```bash
# Install Graphify CLI
pip install graphify-cg
# Or from source
git clone https://github.com/safishamsi/graphify.git
cd graphify
pip install -e .
```
## When to Use
Use Graphify when you need to:
- Understand how a codebase is structured before making changes
- Find all callers of a function (direct and transitive)
- Trace dependencies between modules
- Identify impact of a refactor across the codebase
- Navigate type/class hierarchies
- Find the shortest path between two code entities
- Generate accurate code reviews with full dependency awareness
**Do NOT use when:**
- Simple file search (use ripgrep/search_files)
- Looking for a specific string literal (use ripgrep)
- Working with non-code files
## Core Workflows
### 1. Initialize a Project
```bash
# Initialize Graphify for a project directory
cd /path/to/project
graphify init
# Index the codebase (builds the knowledge graph)
graphify index
# Index with specific languages only
graphify index --lang python,typescript
```
### 2. Query the Knowledge Graph
```bash
# Natural language query
graphify query "What services call database methods?"
# Find all dependencies of a module
graphify deps UserService
# Find all callers of a function
graphify callers main
graphify callers "DatabaseService.save"
# Trace a call chain
graphify trace "APIHandler.process_request" --depth 5
# Find shortest path between two nodes
graphify path "APIHandler" "DatabaseConnection"
```
### 3. Structured Output (for LLM consumption)
```bash
# JSON output — pipe directly to LLM
graphify query "What depends on the auth module?" --format json
# Streaming JSON for large results
graphify deps LargeModule --format json --stream
```
### 4. Git Integration
```bash
# Enable auto-refresh on git operations
graphify hooks install
# Manual refresh after changes
graphify refresh
```
## Hermes Agent Integration
### Code Understanding Before Changes
Before modifying code, use Graphify to understand the full picture:
```bash
# 1. Understand what you're touching
graphify deps "module_name" --format json
# 2. Find all callers (who depends on this?)
graphify callers "function_name" --format json
# 3. Check refactoring safety
graphify path "entry_point" "target_module" --format json
```
### Inject into System Prompt
For complex tasks, inject the codebase structure into the agent's context:
```bash
# Get a structural overview
graphify query "Give me a high-level overview of the project structure" --format json
```
Then include the JSON output in the system message or user message to ground the agent's understanding.
### With Test-Driven Development
Use Graphify to understand test impact:
```bash
# What code does this test exercise?
graphify callers "TestAuth.test_login" --depth 3 --format json
# What tests cover this function?
graphify callers "AuthService.authenticate" --format json | grep test
```
### With Systematic Debugging
Trace bugs through the call chain:
```bash
# Where does this error originate?
graphify trace "APIHandler.handle_request" --format json
# What calls this failing function?
graphify callers "PaymentService.charge" --depth 5
```
## Tips
- **Run `graphify index` after pulling changes** — the graph goes stale otherwise
- **Use `--format json` for agent integration** — structured output is easier to consume
- **Combine with `search_files`** — use Graphify for structure, ripgrep for content
- **Cache is project-local** — each project has its own graph, no cross-contamination
- **Large repos?** Index only the directories you're working in: `graphify index --path src/core`
## Troubleshooting
| Problem | Solution |
|---------|----------|
| "graphify: command not found" | `pip install graphify-cg` |
| "No graph found" | Run `graphify init && graphify index` first |
| Stale results | Run `graphify refresh` |
| Language not supported | Check `graphify --help` for supported list |
| Slow indexing | Use `--path` to limit scope |

View File

@@ -0,0 +1,246 @@
"""Tests for Browser Use provider — anti-detect, CAPTCHA, profiles, persistence."""
from unittest.mock import patch, MagicMock
import pytest
class TestBrowserUseProviderConfig:
"""Test configuration resolution for Browser Use provider."""
def test_not_configured_without_key(self, monkeypatch):
monkeypatch.delenv("BROWSER_USE_API_KEY", raising=False)
from tools.browser_providers.browser_use import BrowserUseProvider
provider = BrowserUseProvider()
assert provider.is_configured() is False
def test_configured_with_api_key(self, monkeypatch):
monkeypatch.setenv("BROWSER_USE_API_KEY", "test-key-123")
from tools.browser_providers.browser_use import BrowserUseProvider
provider = BrowserUseProvider()
assert provider.is_configured() is True
def test_provider_name(self):
from tools.browser_providers.browser_use import BrowserUseProvider
assert BrowserUseProvider().provider_name() == "Browser Use"
class TestBrowserUseFeatures:
"""Test feature configuration and payload assembly."""
def _make_provider(self, monkeypatch, **env):
defaults = {
"BROWSER_USE_API_KEY": "test-key",
"BROWSER_USE_ANTI_DETECT": "true",
"BROWSER_USE_CAPTCHA_SOLVING": "true",
"BROWSER_USE_PERSISTENT_LOGIN": "false",
"BROWSER_USE_PROFILE_NAME": "",
"BROWSER_USE_PROXY_COUNTRY": "us",
}
defaults.update(env)
for k, v in defaults.items():
if v is None:
monkeypatch.delenv(k, raising=False)
else:
monkeypatch.setenv(k, v)
# Reimport to pick up new env values
import importlib
import tools.browser_providers.browser_use as mod
importlib.reload(mod)
return mod.BrowserUseProvider()
def test_features_all_enabled(self, monkeypatch):
provider = self._make_provider(
monkeypatch,
BROWSER_USE_ANTI_DETECT="true",
BROWSER_USE_CAPTCHA_SOLVING="true",
BROWSER_USE_PERSISTENT_LOGIN="true",
BROWSER_USE_PROFILE_NAME="my-profile",
)
features = provider._build_features()
assert features["anti_detect"] is True
assert features["captcha_solving"] is True
assert features["persistent_login"] is True
assert features["profile"] is True
assert features["proxy_country"] == "us"
def test_features_all_disabled(self, monkeypatch):
provider = self._make_provider(
monkeypatch,
BROWSER_USE_ANTI_DETECT="false",
BROWSER_USE_CAPTCHA_SOLVING="false",
BROWSER_USE_PERSISTENT_LOGIN="false",
BROWSER_USE_PROFILE_NAME="",
)
features = provider._build_features()
assert features["anti_detect"] is False
assert features["captcha_solving"] is False
assert features["persistent_login"] is False
assert features["profile"] is False
def test_payload_includes_anti_detect(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_ANTI_DETECT="true")
payload = provider._build_create_payload(managed_mode=False)
assert payload.get("antiDetect") is True
def test_payload_excludes_anti_detect_when_disabled(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_ANTI_DETECT="false")
payload = provider._build_create_payload(managed_mode=False)
assert "antiDetect" not in payload
def test_payload_includes_captcha(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_CAPTCHA_SOLVING="true")
payload = provider._build_create_payload(managed_mode=False)
assert payload.get("captchaSolving") is True
def test_payload_includes_persistent_login(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_PERSISTENT_LOGIN="true")
payload = provider._build_create_payload(managed_mode=False)
assert payload.get("keepCookies") is True
def test_payload_includes_profile(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_PROFILE_NAME="work-account")
payload = provider._build_create_payload(managed_mode=False)
assert payload.get("profileName") == "work-account"
def test_payload_includes_proxy_country(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_PROXY_COUNTRY="de")
payload = provider._build_create_payload(managed_mode=False)
assert payload.get("proxyCountryCode") == "de"
def test_managed_mode_payload(self, monkeypatch):
provider = self._make_provider(monkeypatch)
payload = provider._build_create_payload(managed_mode=True)
assert payload["timeout"] == 5
assert payload["proxyCountryCode"] == "us"
def test_empty_profile_excluded(self, monkeypatch):
provider = self._make_provider(monkeypatch, BROWSER_USE_PROFILE_NAME="")
payload = provider._build_create_payload(managed_mode=False)
assert "profileName" not in payload
class TestBrowserUseSessionCreation:
"""Test session creation with mocked HTTP."""
def _make_provider(self, monkeypatch):
monkeypatch.setenv("BROWSER_USE_API_KEY", "test-key")
import importlib
import tools.browser_providers.browser_use as mod
importlib.reload(mod)
return mod.BrowserUseProvider()
def test_create_session_returns_features(self, monkeypatch):
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.ok = True
mock_response.json.return_value = {
"id": "sess-123",
"cdpUrl": "wss://cdp.browser-use.com/session/123",
}
mock_response.headers = {}
with patch("tools.browser_providers.browser_use.requests.post", return_value=mock_response):
result = provider.create_session("task-1")
assert result["bb_session_id"] == "sess-123"
assert result["features"]["browser_use"] is True
assert result["features"]["anti_detect"] is True
assert result["features"]["captcha_solving"] is True
def test_create_session_sends_payload(self, monkeypatch):
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.ok = True
mock_response.json.return_value = {"id": "sess-456", "cdpUrl": "wss://cdp.example.com"}
mock_response.headers = {}
captured_payload = {}
def mock_post(url, **kwargs):
captured_payload.update(kwargs.get("json", {}))
return mock_response
with patch("tools.browser_providers.browser_use.requests.post", side_effect=mock_post):
provider.create_session("task-2")
assert captured_payload.get("antiDetect") is True
assert captured_payload.get("captchaSolving") is True
def test_create_session_with_profile(self, monkeypatch):
monkeypatch.setenv("BROWSER_USE_PROFILE_NAME", "shared-team")
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.ok = True
mock_response.json.return_value = {"id": "sess-789", "cdpUrl": "wss://cdp.example.com"}
mock_response.headers = {}
captured_payload = {}
def mock_post(url, **kwargs):
captured_payload.update(kwargs.get("json", {}))
return mock_response
with patch("tools.browser_providers.browser_use.requests.post", side_effect=mock_post):
provider.create_session("task-3")
assert captured_payload.get("profileName") == "shared-team"
def test_create_session_error_raises(self, monkeypatch):
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.ok = False
mock_response.status_code = 401
mock_response.text = "Unauthorized"
with patch("tools.browser_providers.browser_use.requests.post", return_value=mock_response):
with pytest.raises(RuntimeError, match="Failed to create Browser Use session"):
provider.create_session("task-err")
class TestBrowserUseSessionCleanup:
"""Test session cleanup."""
def _make_provider(self, monkeypatch):
monkeypatch.setenv("BROWSER_USE_API_KEY", "test-key")
import importlib
import tools.browser_providers.browser_use as mod
importlib.reload(mod)
return mod.BrowserUseProvider()
def test_close_session_success(self, monkeypatch):
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.status_code = 200
with patch("tools.browser_providers.browser_use.requests.patch", return_value=mock_response):
result = provider.close_session("sess-123")
assert result is True
def test_close_session_failure(self, monkeypatch):
provider = self._make_provider(monkeypatch)
mock_response = MagicMock()
mock_response.status_code = 500
mock_response.text = "Internal Server Error"
with patch("tools.browser_providers.browser_use.requests.patch", return_value=mock_response):
result = provider.close_session("sess-bad")
assert result is False
def test_emergency_cleanup_no_error_on_failure(self, monkeypatch):
provider = self._make_provider(monkeypatch)
with patch("tools.browser_providers.browser_use.requests.patch", side_effect=Exception("network down")):
# Should not raise
provider.emergency_cleanup("sess-xyz")

View File

@@ -1,4 +1,8 @@
"""Browser Use cloud browser provider."""
"""Browser Use cloud browser provider.
Enhanced with anti-detect profiles, CAPTCHA solving, persistent logins,
and profile management per Issue #262.
"""
import logging
import os
@@ -21,6 +25,17 @@ _DEFAULT_MANAGED_TIMEOUT_MINUTES = 5
_DEFAULT_MANAGED_PROXY_COUNTRY_CODE = "us"
# ---------------------------------------------------------------------------
# Config helpers
# ---------------------------------------------------------------------------
def _env_bool(key: str, default: bool = False) -> bool:
val = os.environ.get(key)
if val is None:
return default
return val.lower() in ("1", "true", "yes", "on")
def _get_or_create_pending_create_key(task_id: str) -> str:
with _pending_create_keys_lock:
existing = _pending_create_keys.get(task_id)
@@ -61,7 +76,28 @@ def _should_preserve_pending_create_key(response: requests.Response) -> bool:
class BrowserUseProvider(CloudBrowserProvider):
"""Browser Use (https://browser-use.com) cloud browser backend."""
"""Browser Use (https://browser-use.com) cloud browser backend.
Supports anti-detect profiles, CAPTCHA solving, persistent logins,
and named profile management. Configuration via env vars:
- ``BROWSER_USE_API_KEY`` — direct API key
- ``BROWSER_USE_ANTI_DETECT`` — enable anti-detect fingerprinting (default: true)
- ``BROWSER_USE_CAPTCHA_SOLVING`` — enable CAPTCHA auto-solving (default: true)
- ``BROWSER_USE_PERSISTENT_LOGIN`` — persist cookies across sessions (default: false)
- ``BROWSER_USE_PROFILE_NAME`` — named profile for cookie jar isolation
- ``BROWSER_USE_PROXY_COUNTRY`` — proxy country code override (default: us)
"""
# Feature config snapshot — read once at import so runtime env mutation
# cannot silently change an in-flight session's capabilities.
_cfg_anti_detect: bool = _env_bool("BROWSER_USE_ANTI_DETECT", True)
_cfg_captcha: bool = _env_bool("BROWSER_USE_CAPTCHA_SOLVING", True)
_cfg_persistent: bool = _env_bool("BROWSER_USE_PERSISTENT_LOGIN", False)
_cfg_profile: str = os.environ.get("BROWSER_USE_PROFILE_NAME", "")
_cfg_proxy_country: str = os.environ.get(
"BROWSER_USE_PROXY_COUNTRY", _DEFAULT_MANAGED_PROXY_COUNTRY_CODE
)
def provider_name(self) -> str:
return "Browser Use"
@@ -106,6 +142,56 @@ class BrowserUseProvider(CloudBrowserProvider):
raise ValueError(message)
return config
# ------------------------------------------------------------------
# Feature / payload assembly
# ------------------------------------------------------------------
def _build_features(self) -> Dict[str, Any]:
"""Return a dict describing which features are active."""
return {
"browser_use": True,
"anti_detect": self._cfg_anti_detect,
"captcha_solving": self._cfg_captcha,
"persistent_login": self._cfg_persistent,
"profile": bool(self._cfg_profile),
"proxy_country": self._cfg_proxy_country,
}
def _build_create_payload(self, managed_mode: bool) -> Dict[str, Any]:
"""Build the session creation payload with all configured features."""
payload: Dict[str, Any] = {}
if managed_mode:
payload["timeout"] = _DEFAULT_MANAGED_TIMEOUT_MINUTES
payload["proxyCountryCode"] = self._cfg_proxy_country
elif self._cfg_proxy_country:
payload["proxyCountryCode"] = self._cfg_proxy_country
# Anti-detect fingerprinting — Browser Use v3 uses
# the antiDetect field to enable browser fingerprint spoofing
# so sites see a real human browser, not an automation tool.
if self._cfg_anti_detect:
payload["antiDetect"] = True
# CAPTCHA solving — Browser Use handles reCAPTCHA, hCaptcha,
# Cloudflare Turnstile, and custom CAPTCHAs automatically.
if self._cfg_captcha:
payload["captchaSolving"] = True
# Persistent login — preserves cookies across sessions when
# a named profile is provided, enabling sites like Gmail, Twitter,
# etc. to stay logged in between agent runs.
if self._cfg_persistent:
payload["keepCookies"] = True
# Named profile — isolates cookie jar and browser state under
# a specific profile name. Multiple agents can share a profile
# or use isolated profiles for different accounts.
if self._cfg_profile:
payload["profileName"] = self._cfg_profile
return payload
# ------------------------------------------------------------------
# Session lifecycle
# ------------------------------------------------------------------
@@ -125,17 +211,7 @@ class BrowserUseProvider(CloudBrowserProvider):
if managed_mode:
headers["X-Idempotency-Key"] = _get_or_create_pending_create_key(task_id)
# Keep gateway-backed sessions short so billing authorization does not
# default to a long Browser-Use timeout when Hermes only needs a task-
# scoped ephemeral browser.
payload = (
{
"timeout": _DEFAULT_MANAGED_TIMEOUT_MINUTES,
"proxyCountryCode": _DEFAULT_MANAGED_PROXY_COUNTRY_CODE,
}
if managed_mode
else {}
)
payload = self._build_create_payload(managed_mode)
response = requests.post(
f"{config['base_url']}/browsers",
@@ -158,7 +234,14 @@ class BrowserUseProvider(CloudBrowserProvider):
session_name = f"hermes_{task_id}_{uuid.uuid4().hex[:8]}"
external_call_id = response.headers.get("x-external-call-id") if managed_mode else None
logger.info("Created Browser Use session %s", session_name)
logger.info(
"Created Browser Use session %s [anti_detect=%s captcha=%s persistent=%s profile=%s]",
session_name,
self._cfg_anti_detect,
self._cfg_captcha,
self._cfg_persistent,
self._cfg_profile or "(none)",
)
cdp_url = session_data.get("cdpUrl") or session_data.get("connectUrl") or ""
@@ -166,7 +249,7 @@ class BrowserUseProvider(CloudBrowserProvider):
"session_name": session_name,
"bb_session_id": session_data["id"],
"cdp_url": cdp_url,
"features": {"browser_use": True},
"features": self._build_features(),
"external_call_id": external_call_id,
}