Skills

Lightweight framework for comparing search providers, RAG pipelines, and retrieval quality. Build golden sets, run LLM-as-Judge scoring.

Automatically determine whether to parallelize agent tasks or run single-threaded. Evaluates complexity, decomposability, cost and dependencies to recommend agent count and merge strategy.

Real-time knowledge graph memory for AI agents. Writes every conversation turn to Zep Cloud, retrieves relevant facts before each response. Zero extra model compute.

Recursive Language Model pattern for processing arbitrarily long documents. Treats input as external environment with search, slice, and parallel reading of specific regions.

Monitors your active positions against a Zep knowledge graph. Daily contradiction checks flag when new evidence supports or contradicts your theses. Catches stale assumptions.

Run any task through multiple AI models and cross-check for errors, hallucinations, and blind spots. Catches what single models miss.

Scans your Zep knowledge graph for recurring entity clusters and converging patterns. Finds when 3+ sources confirm the same signal. Surfaces trends before they become obvious.

Observe-and-report monitoring for any AI agent. Auto-discovers identity, runs diagnostics, flags issues, and produces fix prompts. Never auto-fixes. Two-phase design.

Design anti-pattern detection for HTML, CSS, and frontend code. Catches accessibility issues, spacing problems, and visual inconsistencies before they ship.

Hardened autonomous agents for financial workflows. Audited (DeepSeek V4 Pro + Grok 4.3). Four patches applied: bash removed, MCP security hardened, and more.

npm for agent skills. Install, update, and remove SKILL.md files. Built by Vercel. Distribution marketplace for agent capabilities.

MCP server that gives agents up-to-date library docs and code examples. Stops hallucinated APIs by grounding agents in real documentation.

AI agent instructions for service and maintenance. Diagnose, archive, maintenance refactor, and optimization patterns for keeping agents healthy long-term.

Convention for placing agent instructions in AGENTS.md files. Zero dependencies, 270 bytes. The standard way to give any AI agent context about your project.

Control web pages with the OpenClaw browser tool. Multi-step flows, login checks, tab management, and recovery from stale refs or timeouts.

Audit, trim, and optimize agent context without losing recall. Six-phase migrate-verify-slim sequence. Reduces token load 75-90% while keeping all content searchable.

Self-improvement loop for AI agents. Trace every completed task, judge with a different model, diagnose patterns, and apply concrete fixes to skills and memory. Compound optimization over time.