agent▌

AI agent architecture for anticipating needs, surviving context loss, and continuous self-improvement. \n \n Implements Write-Ahead Logging (WAL) Protocol to capture corrections, decisions, and preferences before responding, ensuring critical details persist across sessions \n Three-tier memory system (SESSION-STATE.md, daily logs, curated MEMORY.md) with Working Buffer for context survival during compaction and Compaction Recovery for seamless resumption \n Proactive behavior patterns including

Playwright-based browser automation with element refs for AI agents, supporting navigation, interaction, screenshots, and video recording. \n \n Provides 6 core functions: open (navigate with config), snapshot (refresh element refs), interact (click/fill/drag/upload/scroll), screenshot, execute (JavaScript), and close \n Element interaction uses simple @e ref system that invalidates after navigation, requiring re-snapshot calls to maintain accurate selectors \n Supports video recording with opti

Skill by ara.so — Daily 2026 Skills collection.

Universal self-improving agent that learns from all skill experiences using multi-memory architecture. \n \n Implements semantic, episodic, and working memory to extract patterns, abstract insights, and continuously evolve skill guidance across the codebase \n Auto-triggers on skill completion, errors, and session events via hooks-based integration; detects and corrects inaccurate guidance with traceable evolution markers \n Prioritizes updates across 10+ skill categories (PRD planning, architec

Drop-in React/Next.js agent component with runtime, tools, streaming, and human-in-the-loop approvals built in. \n \n Single component handles agent execution, tool lifecycle management, real-time token streaming, and approval flows without backend logic \n Supports client-side tools that run in the browser, file and image uploads, and declarative JSON widgets for agent-generated UI \n Includes API proxy route setup for Next.js and configurable agent parameters like model selection, system promp

AI-powered autonomous extraction of structured data from complex multi-page websites. \n \n Navigates sites intelligently to locate and extract data, returning results as JSON with optional schema validation \n Supports custom JSON schemas for predictable structured output, or freeform extraction when schema is not provided \n Offers two model tiers (spark-1-mini and spark-1-pro) with credit limits and optional waiting for inline results \n Best suited for multi-page extraction tasks; use simple

Establish AI agent environment policies, security guardrails, and team configuration standards. \n \n Configure project description files as AI manuals with tech stack, coding standards, and DO NOT rules; use /init for auto-generation from codebase analysis \n Set up security hooks to block dangerous commands (rm -rf, sudo, curl | sh) and auto-approve only safe operations via PreToolUse and PostToolUse events \n Define skills, slash commands, and plugins with token efficiency in mind; skills loa

Comprehensive evaluation framework for designing, building, and monitoring AI agent performance across coding, conversational, research, and computer-use agents. \n \n Covers three grader types (code-based, model-based, human) with trade-offs and best practices for each agent category \n Provides an 8-step roadmap from initial task creation through production monitoring, including environment isolation, outcome-focused grading, and saturation detection \n Includes benchmarks for major agent type

Conversational AI video creation agent that plans, generates, and delivers finished videos from natural language descriptions. \n \n Supports short-form video output (5–60 seconds) in three aspect ratios: 16:9, 9:16, and 1:1, suitable for YouTube, TikTok, Instagram, and other platforms \n Accepts reference materials including product photos, brand assets, style examples, and audio files to guide creative direction and visual consistency \n Engages in multi-turn dialogue, asking clarifying questi

Refactor bloated agent instruction files into organized, linked documentation using progressive disclosure. \n \n Analyzes existing AGENTS.md, CLAUDE.md, or similar files to identify contradictions, extract essentials, and categorize remaining instructions across 3–8 focused topic files \n Keeps root file minimal (under 50 lines) with only universal project info and links to detailed guidelines organized by topic (TypeScript, testing, code style, git workflow, architecture) \n Flags redundant, v