Skill UI

This skill provides a systematic, four-phase workflow for diagnosing and recovering from AI agent failures. It guides the agent to first capture the failure state, diagnose the root cause (e.g., loop, context overflow, state drift), apply minimal containment recovery actions, and finally generate a detailed introspection report, minimizing the need for human intervention.

This skill provides comprehensive interaction with the DeerFlow AI agent platform via its robust HTTP API. It is designed for managing complex, multi-step AI workflows, including deep academic research, code execution, web browsing, and file analysis. Users can initiate and continue conversation threads, manage agent memory, upload documents (PDF, etc.), and delegate complex research tasks to the autonomous agent system.

A cultural pressure‑style skill that forces the assistant to repeatedly push past failures, think systemically, validate every change, and proactively hunt down related issues until end‑to‑end ownership is demonstrated.

Provides comprehensive patterns and techniques for securing AI agent systems. Learn how to enforce governance policies, including tool allow/block lists, content filtering, rate limiting, and human approval workflows. Essential for building robust, compliant, and trustworthy agents in multi-agent or production environments that interact with external APIs and sensitive data.

This skill provides a comprehensive framework for optimizing, improving, and debugging Large Language Model (LLM) prompts. It leverages production trace data, system evaluations, and human annotations to extract prompts from spans. Users can run a data-driven optimization loop, ensuring the AI responses meet desired quality standards. Ideal for prompt tuning, system prompt refinement, and improving overall AI output quality.

A comprehensive guide for developers to build advanced, declarative agents for Microsoft 365 Copilot. This skill details the integration process using a Model Context Protocol (MCP) server, allowing the agent to connect with external systems, select tools, and handle complex authentication (OAuth 2.0/SSO). You will learn to generate key JSON manifests required for deployment.

Phoenix Evals enables developers to build and run evaluators for AI and LLM applications. It supports both code-based and LLM-based evaluation methods, including RAG systems. Users can create datasets, run experiments, validate evaluator accuracy, and implement production guardrails. Available in Python and TypeScript, it helps ensure model performance and reliability through systematic testing and human validation.

A continuous file-based planning system for multi-step AI agents. Maintains task_plan.md, findings.md, and progress.md files on disk, injecting project-specific planning context into lifecycle hooks. Auto-restoration reads only project planning files. Can inspect local agent session metadata with session-catchup.py --metadata, or replay limited excerpts with --replay. Optional controlled mode requests continuation only when supported by host. Never executes declared Markdown commands. No network data upload path. Used for research or work requiring 5+ tool calls.

This tool generates a complete, structured TypeSpec declarative agent definition tailored for Microsoft 365 Copilot. It systematically creates necessary components including detailed role instructions, conversation starters, and a comprehensive list of required capabilities (e.g., WebSearch, OneDrive, Teams, CodeInterpreter). This ensures the agent is fully functional and ready for deployment within the M365 ecosystem.

A structured methodology for precisely identifying specific geographic locations or historical landmarks. This technique systematically narrows the search scope, moving from broad regions (countries/former states) down through administrative divisions, city centers, and finally to specific monuments or structures. It requires utilizing diverse clues like historical events, census data, and distance constraints, while critically verifying assumptions at each step to ensure accuracy.

Acontext serves as a robust memory layer for production-grade AI agents. It provides persistent sessions, disk-based storage, and sophisticated learning spaces that distill complex conversations into reusable, structured skills. Users can manage the entire system—from project setup and login to skill uploading—via a powerful Command Line Interface (CLI) and API, significantly enhancing agents' contextual understanding and long-term knowledge retention.

Evolver is a comprehensive self-evolution engine designed for autonomous AI agents. It analyzes complex runtime history, identifies performance bottlenecks, and autonomously generates and applies improvements. Communication is architecturally confined to a local Proxy Mailbox, ensuring secure and reliable interaction with the EvoMap Hub for asset management and task acquisition. Ideal for developing highly adaptive, continuously improving intelligent systems.