Skills & Tools

OpenClaw v2026.7.2-beta.2 introduces remote coding sessions on cloud workers, native automation and mobile capabilities, safer channel operations, guided Control UI setup, improved gateway and session recovery, and enhanced Linux packaging. Key improvements include external gateway supervision, ClickClack integration, Skill Workshop enhancements, and numerous Control UI refinements for better user experience and reliability.

crewAI 1.15.3 introduces enhanced execution interception points, declarative flow support in TUI, and organizational API improvements. Key bug fixes address hook execution, tool caching, token usage reporting, and history management. The release improves developer experience through better hook documentation and more granular control over agent execution boundaries.

Martin is an LLM-powered voice assistant (YC S23) that improves upon Siri by offering deeper app integrations, better personalization through conversation memory, and multi-channel access (voice, SMS, WhatsApp, email). Key features include calendar management, reminder setting, autonomous texting with contacts on behalf of users, and daily briefings. The platform addresses legacy voice assistant limitations through optimized latency, custom voice activity detection, and sophisticated memory systems that learn user context from conversations.

Baseplate is a unified backend-as-a-service platform for LLM applications that simplifies the infrastructure needed to build AI-powered products. It provides integrated data management, vector embeddings, prompt versioning, and deployment capabilities through a single API, eliminating the need for teams to build separate datasource integrations, vector databases, and storage solutions. The platform features a multimodal database, spreadsheet-style editing interface, hybrid search capabilities, and an App Builder for prompt iteration and context injection.

Onyx is an open-source chat UI (YC W24) that works with any LLM and provides enterprise-grade features like RAG, web search, MCP, and deep research. Built by the creators of Danswer, the team pivoted after realizing users wanted secure, self-hosted chat interfaces with world-class UX rather than just search. The platform supports RBAC, SSO, and on-prem hosting, with key innovations in context management and model-agnostic tool handling that enable reliable multi-model deployments at scale.

Continue is an open-source coding autopilot for VS Code that replaces manual ChatGPT copy-pasting with an integrated LLM assistant. It emphasizes transparency, hackability, and user control by supporting any LLM (local or cloud), collecting development data locally, and offering extensive customization through slash commands, context sources, and policies. The platform enables developers to maintain privacy while leveraging AI assistance and automatically gather data to improve their own LLM models.

Forge is an open-source reliability layer for self-hosted LLM tool-calling that dramatically improves agentic task performance on consumer hardware. It uses domain-agnostic guardrails (retry nudges, step enforcement, error recovery, VRAM management) to boost an 8B model from 53% to 99% accuracy on multi-step workflows without model changes. The framework addresses the compounding failure problem in local models—90% per-step accuracy yields only 59% success on 5-step tasks—and peer-reviewed results show an 8B model with Forge (99.3%) outperforming Claude Sonnet without guardrails (87.2%).

OpenClaw v2026.7.2-beta.1 introduces remote coding sessions on cloud workers, native automation and mobile capabilities, safer channel operations, guided Control UI setup, improved gateway and session recovery, and new Linux packaging options. The release encompasses 2425 merged PRs with major enhancements to session management, coding agents, security boundaries, and reliability across multiple platforms including mobile, desktop, and headless systems.

OpenClaw v2026.7.1 introduces major platform enhancements including new AI models (Claude Sonnet 5, Mythos 5, Meta Muse Spark 1.1), a redesigned Control UI with sessions-first navigation, conversational onboarding via Crestodian agent-loop, offline/spoken mobile chat capabilities, and improved session organization with auto-generated titles. The release also strengthens Telegram/Codex continuity, adds startup/upgrade recovery mechanisms, and expands provider support with ClawRouter for dynamic model discovery and budget management.

crewAI 1.15.2 introduces dynamic LLM model loading in the crew wizard, inline skill definitions, and enhanced Flow capabilities including templated action inputs and a new Flow Definition authoring skill. The release includes message setup/feedback handling in AgentExecutor, repository agents support, and a stream frame protocol for flows. Multiple bug fixes address model-catalog caching, pip-audit vulnerabilities, and flow input resolution.

Google ADK Python v2.4.0 introduces significant enhancements including mTLS support across multiple tools, OpenAI Responses API integration, Managed Agents API backing, and streaming improvements for thoughts and media. The release adds workflow-as-tool capabilities, enhanced database session management, and improved analytics with thinking/tool-use token tracking. Multiple bug fixes address security concerns, path traversal prevention, and SDK compatibility updates.

Claude Code v2.1.202 introduces dynamic workflow size configuration, enhanced OpenTelemetry telemetry attributes for workflow tracing, and fixes 17+ critical bugs spanning interactive sessions, Remote Control integration, voice dictation, and workflow parsing. Key improvements include better sign-in URL handling over SSH, improved workflow agent list UI, and enhanced MCP error messaging. The release also optimizes session resumption performance in repositories with many git worktrees.

Claude Code v2.1.199 is a maintenance release focused on reliability and user experience improvements. Key fixes address SSL certificate handling, streaming response preservation, subagent error reporting, background agent stability on Linux and macOS, and UI/UX issues like progress indicators and agent panel display. The release also improves retry logic for transient server errors and fixes several edge cases in session management and tool routing.

Claude Code v2.1.198 introduces Claude in Chrome as generally available, adds background agent notifications and a new `/dataviz` skill for chart design, and expands upstream provider support with Claude Platform on AWS. The release includes significant improvements to agent workflows, including automatic PR creation for code work, better error handling with transient retry logic, and inheritance of session configurations by subagents. Multiple bug fixes address agent lifecycle management, network resilience, UI rendering, and authentication token refresh.

This cookbook demonstrates how to reproduce Claude's published agentic search benchmark scores (DeepSearchQA, BrowseComp) using the Messages API with programmatic tool calling, server-side compaction, and task budgets. The key is proper harness configuration—API parameters that become critical for agents running 30+ tool calls across hundreds of thousands of tokens. By following this guide, you'll build an agentic search loop that matches Claude's official benchmark performance and understand why each configuration choice matters for long-horizon tasks.

crewAI 1.15.2a1 release introduces inline skill definitions, stream frame protocol for flows, and enhanced type support in CrewDefinition. Key improvements include repointing template commands to crewAIInc-fde org and adding a generated Flow Definition authoring skill. Bug fixes address documentation navigation issues, with updates to cost limit rules and logging configuration guides.

Claude Code v2.1.195 release introduces environment variable control for mouse interactions, fixes critical issues with hook matchers, voice dictation across multiple platforms, plugin management, and background job persistence. Key improvements include better error messaging for voice mode on Linux, enhanced UI for agent management, and more reliable remote session provisioning. This update addresses stability and usability concerns across macOS, Linux, and remote environments.

crewAI 1.15.0 introduces significant enhancements to conversational flows and declarative configuration, including unified flow loading, DMN mode support, and CLI TUI integration. The release adds new composite actions (each, single agent, crew actions) to FlowDefinition and improves telemetry tracking for conversational turn usage. Multiple bug fixes address JSON schema handling, credential permissions, token aggregation, and symlink vulnerabilities, while documentation improvements include Datadog integration guides and JSON-first crew project setup.

Claude Code v2.1.186 introduces CLI authentication for MCP servers, enhanced workflow filtering, and improved agent team management. Key fixes address streaming failures after sleep, subagent transcript issues, and permission prompt handling. The release also improves skill metadata flexibility, memory management, and background session behavior with better user control over tool permissions.

OpenClaw v2026.6.10-beta.1 is a major release focused on reliability, security, and feature expansion. Key improvements include more stable agent turns and session state management, stronger Codex and approval flows, richer channel delivery across Telegram/Discord/Slack, safer network boundaries, and new CLI workflows. The release also enhances mobile/desktop clients, expands plugin coverage, and fixes critical issues in transcript correctness, security, and provider authentication across 109 merged PRs.