Repository navigation
Releases
Releases · sl4de0day/wide-sdk
Wide SDK 0.4141126
Wide SDK v0.4141126
Highlights
- Real CDP Browser Controller and Automation Tools (
wide_sdk/browser/controller.py,wide_sdk/tools/browser.py):- Added
BrowserControllerand_SimpleCDPClientsupporting 16 browser tools (browser_open_tool,browser_navigate_tool,browser_snapshot_tool,browser_click_tool,browser_type_tool,browser_press_tool,browser_hover_tool,browser_select_tool,browser_scroll_tool,browser_wait_tool,browser_extract_tool,browser_screenshot_tool,browser_console_tool,browser_cookies_tool,browser_eval_tool,browser_close_tool) with@eNaccessibility reference targeting.
- Added
- Database Client, Redaction, and SQL Risk Guards (
wide_sdk/database/,wide_sdk/tools/db.py):- Added
DatabaseClient,DatabaseConnectionSpec,parse_database_target,mask_dsn,redact_rows,is_query_readonly, andassess_sql_riskalongside three agent tools:db_inspect_schema_tool,db_query_tool, anddb_audit_tool.
- Added
- SHA-256 Verified Rules/OSV Sync, Agent Guardrails, and Update Checks (
wide_sdk/security/loader.py,wide_sdk/agent.py,wide_sdk/__init__.py):- Upgraded
RuleLoader(update_rules,load_advisories,_parse_file,_merge_default_agent_rules) withSHA256SUMSverification forrules.jsonandosv-bundle.json, guaranteedwide/ai-no-emojisfallback merging across partial bundles, andnot-matchfalse-positive suppression. - Enforced dynamic
wide-rulesagent guardrails, comprehensive ZWJ/keycap/variation/pictograph no-emojisanitize_output, silent directive adherence inAgent/build_agent_system_prompt, and exportedcheck_for_updates()for version0.4141126. - Verified with
python -m unittest discover -s tests -v(28/28 tests passing).
- Upgraded
Wide SDK 0.3101126
Wide SDK 0.3101126 Release Notes
Autonomous Browser Controller
- Integrated BrowserController (wide_sdk/browser/controller.py) with programmatic HTML & DOM parsing, element interaction indexing, typing, clicking, and viewport snapshot rendering.
- Exposed 6 browser tools (browser_navigate_tool, browser_click_tool, browser_type_tool, browser_extract_tool, browser_screenshot_tool, browser_close_tool) registered by default in Agent.
Dynamic AI Agent Security Rules
- Added RuleLoader.load_agent_rules and get_agent_rules_path (wide_sdk/security/loader.py) supporting dynamic loading from workspace .wide/agent-rules.json, .wide/rules.json, cached rules, and embedded fallback.
- Embedded 8 validated AI security rules (DEFAULT_AGENT_RULES) covering unsafe tool execution, prompt injection, unconstrained eval, unauthenticated MCP registration, unbounded agent loops, context accumulation, prompt secret leaks, and unvalidated patch application.
Wide SDK 0.291126
DecisionCore multi-agent coordination and dynamic security rule loading release.
Wide SDK v0.191126
Wide SDK v0.191126 - Initial Release
Python SDK to build autonomous AI coding and web security agents powered by Wide.
Key Features:
- Autonomous ReAct Agent Loop with context compaction
- Comprehensive 20+ category native SAST engine (SQLi, XSS, SSRF, Command Injection, XXE, Deserialization, etc.)
- Active API endpoint fuzzer and security payload generator
- Multi-provider support (Ollama local offline, Gemini, Claude, DeepSeek, OpenAI)
- Extensible @tool decorator and custom tool interface with automatic schema generation
- Complete type annotations and standard library core architecture
Installation:
pip install wide-sdk