evalmonkey vs Vercel Skills

evalmonkey is an open-source framework for testing how LLM agents degrade under adversarial conditions. You run your agent against 10 standard datasets (GSM8K, ARC, HellaSwag, etc.) pulled automatically from HuggingFace, then apply chaos profiles that introduce realistic failure modes: malformed JSON schemas, artificial latency spikes, 429 rate-limit errors, context-window overflow, and prompt injection payloads. The key output is a degradation delta — evalmonkey shows you exactly how much your agent's accuracy drops under each failure type versus clean inputs. A model that scores 78% on GSM8K normally but drops to 31% when it gets a 429 mid-chain tells you something crucial about its error-recovery behavior that standard benchmarks completely miss. It supports OpenAI, Anthropic (via Bedrock and direct), Azure, GCP, and any Ollama-hosted model. Corbell-AI published this with a clear thesis: agents break in production for infrastructure reasons, not model reasons — and no existing benchmark tests that. evalmonkey was created today (April 17, 2026) and is still at 3 stars, but the core idea is genuinely novel in the evals space.

Vercel Labs Skills is a CLI tool (`npx skills`) that introduces a standardized, portable format for AI agent capabilities. Instead of crafting system prompts project by project, developers install SKILL.md files — YAML-frontmatter instruction sets — globally or per-project, and they work across 40+ coding agents: Claude Code, Cursor, Windsurf, Cline, Continue, and more. The skills ecosystem solves a genuine portability problem: every team that switches tools loses carefully crafted agent instructions. A skill installed once — say, "write tests in Vitest with coverage" or "generate accessible React components" — persists across projects and survives tool migrations. Skills are composable, version-controlled, and shareable via npm or git. Community uptake has been rapid since launch, with a growing registry of skills covering testing, documentation, code review, accessibility, and API design patterns. At 317 GitHub stars on day one, it's the most promising attempt yet at building a cross-agent skill ecosystem — and Vercel's distribution muscle means it's likely to become the de facto standard.