agentsclimarketplace
Sign in

Shelves Context & AI engineering

Evals & benchmarks

944 rows from 487 repositories

Deciding whether the output was actually any good, repeatably.

What Evals & benchmarks skills agree on

508 skill files read, by 499 of the 513 authors on this shelf whose files we hold, 2026-09-06

The middle one of the 63 measured here is ~1.3k tokens long, counted with cl100k_base

Counted by distinct author, so one author publishing three of these counts once. Where a claim sits in fewer files than authors, that is said: a claim held by forty authors across three files is one file people copied, not forty people who agreed. Near-identical wordings are grouped and the other wordings are shown, so the grouping is yours to check.

What they tell the agent to do

  1. Spawn with-skill and baseline runs in the same turn31 of 499 in 24 filesalso worded as Spawn with-skill and baseline runs simultaneously; Launch with-skill and baseline runs together in one turn
  2. Keep SKILL.md under 500 lines31 of 499 in 24 filesalso worded as Keep the SKILL.md body under 500 lines; Keep SKILL.md under 600 lines using references
  3. Draft assertions while test runs are in progress31 of 499 in 24 filesalso worded as Draft assertions while runs are in progress; Draft assertions while the runs execute
  4. Compare against the baseline after changes31 of 499 in 13 filesalso worded as compare current results against a baseline; compare results against the saved baseline
  5. Define evals before coding26 of 499 in 17 filesalso worded as Define evals before writing code; Define eval criteria before writing code
  6. Run evals frequently during development25 of 499 in 16 filesalso worded as Run evals frequently; Run evals continuously during development
  7. Keep evals fast24 of 499 in 15 files
  8. Version evals with code24 of 499 in 15 filesalso worded as Version evals with the code; Version-control evals alongside code
  9. Generate the eval viewer before evaluating outputs yourself24 of 499 in 17 filesalso worded as Launch the eval viewer before reviewing outputs yourself; Generate the eval viewer before reviewing outputs yourself
  10. Generate an eval report after runs24 of 499 in 15 filesalso worded as Generate an eval report after running; Generate a full eval report after runs
  11. Track pass@k metrics over time22 of 499 in 14 filesalso worded as Track pass@k over time; Track pass@k results over time
  12. Save a baseline before making changes21 of 499 in 9 filesalso worded as save first-run results as the baseline; Capture a baseline before making changes
  13. Flag security checks for human review19 of 499 in 10 filesalso worded as Require human review for security checks; Flag security-relevant changes for human review
  14. Track regressions with each change18 of 499 in 9 filesalso worded as Track regressions with every change
  15. Measure Core Web Vitals on each target URL18 of 499 in 6 filesalso worded as measure Core Web Vitals on each page

What they tell it not to do

  1. Do not use other testing skills29 of 499 in 22 filesalso worded as Do not use dedicated testing skills; Don't use other testing skills like /skill-test
  2. Do not write custom HTML for the viewer26 of 499 in 19 filesalso worded as Don't write custom HTML for the viewer; Do not write custom HTML for the eval viewer
  3. Never fully automate security checks25 of 499 in 16 filesalso worded as Do not fully automate security checks
  4. Do not create misleading or malicious skills20 of 499 in 13 filesalso worded as Don't create misleading or malicious skills; Never create misleading or malicious skills
  5. Do not create workspace directories upfront19 of 499 in 13 filesalso worded as Don't create all workspace directories upfront; Do not create all workspace directories upfront
  6. Do not force assertions onto subjective outputs14 of 499 in 8 filesalso worded as Don't force assertions onto subjective outputs; Never write subjective assertions
  7. Do not make negative eval queries obviously irrelevant13 of 499 in 6 filesalso worded as Don't make negative trigger queries obviously irrelevant; Don't make negative eval queries obviously irrelevant
  8. Do not create skills containing malware or exploit code12 of 499also worded as Do not include malware or exploit code in skills; Never include malware or exploit code in skills
  9. Do not write assertions when first saving test cases11 of 499 in 10 filesalso worded as Don't draft assertions before running tests; Don't write assertions when first saving test prompts
  10. Do not edit model cards or model-index10 of 499 in 4 files

What they expect to be installed

  1. grep51 of 499 in 39 files
  2. subagents31 of 499 in 24 files
  3. python330 of 499
  4. git29 of 499 in 28 files
  5. generate_review.py28 of 499 in 21 files
  6. npm run build27 of 499 in 16 files
  7. npm test25 of 499 in 16 files
  8. pytest22 of 499 in 19 files
  9. uv21 of 499 in 13 files
  10. /eval define21 of 499 in 12 files

What they ask it to produce

  1. grading.json per run38 of 499 in 31 filesalso worded as grading.json; grading.json per eval
  2. benchmark.json and benchmark.md35 of 499 in 28 filesalso worded as benchmark.json; benchmark.md
  3. timing.json per run34 of 499 in 27 filesalso worded as timing.json; timing.json with run statistics
  4. Eval report33 of 499 in 25 filesalso worded as Full eval report; eval run report
  5. eval_metadata.json per test case33 of 499 in 26 filesalso worded as eval_metadata.json; evals.json test case file
  6. Packaged .skill file31 of 499 in 24 filesalso worded as .skill package; .skill package file
  7. SKILL.md31 of 499 in 24 filesalso worded as SKILL.md draft; Draft SKILL.md
  8. Eval definition file26 of 499 in 17 filesalso worded as Eval definition markdown file; eval specification file

When Evals & benchmarks authors say to reach for one

The situations these authors wrote into their own files, counted out of the same 499 authors, with the skills that name each one

How Evals & benchmarks skills are built

902 skill directories by 448 authors, read from their repositories’ own file trees 2026-08-05

The middle bundle among those shipping files is 5 files, 30.9 KB beside SKILL.md

Counted by distinct author, same as above, so one author publishing forty template copies counts once. SKILL.md itself is not counted as a file, so a single-file skill is one where that file is the whole skill.

The shape

  1. SKILL.md is the whole skill143 of 448 authors, 400 of 902 skills
  2. files ship beside it305 of 448 authors, 502 of 902 skills
  3. executable scripts ship inside199 of 448 authors, 270 of 902 skills

The folders they converge on

  1. references/226 of 448 authors, 351 of 902 skills
  2. scripts/178 of 448 authors, 246 of 902 skills
  3. agents/127 of 448 authors, 178 of 902 skills
  4. assets/121 of 448 authors, 186 of 902 skills
  5. eval-viewer/100 of 448 authors, 126 of 902 skills
  6. evals/34 of 448 authors, 67 of 902 skills

  • Skill creator

    anthropics/skills/skills/skill-creator Skill

    in 1 stackno license168,934★ repo

    Public repository for Agent Skills

  • Golang benchmark

    samber/cc-skills-golang/skills/golang-benchmark Skill

    2,949★ repo

    🧑‍🎨 A collection of Golang agentic skills that works

  • Woz benchmark

    WithWoz/wozcode-plugin/codex/wozcode/skills/woz-benchmark Skill

    no license201★ repo

    WOZCODE plugin for Claude Code

  • Genesis evals

    danielmeppiel/genesis/dev/skills/genesis-evals Skill

    61★ repo

    Markdown that steers an LLM is code. Genesis is the architectural layer for designing multi-agent, multi-skill systems -- with named patterns, contracts, and substrate portability, before you write them.

  • Writing prds executable

    liqiongyu/lenny_skills_plus/samples/writing-prds-executable Skill

    51★ repo

    86 agent-executable skill packs converted from RefoundAI’s Lenny skills (unofficial). Works with Codex + Claude Code.

  • Research proof

    tonyblu331/research-proof/plugins/research-proof-plugin/skills/research-proof Skill

    45★ repo

    Pressure-test research claims with falsifiable evidence plans, adversarial checks, frozen verifiers, and proof ledgers.

  • Add llm evals

    ContextJet-ai/awesome-llm-observability/skills/add-llm-evals Skill

    no license29★ repo

    50+ curated LLM observability tools PLUS 26 Agent Skills (several with runnable, unit-tested scripts) to build, evaluate, debug, secure & monitor reliable LLM apps. Tracing, evals, guardrails, LLMOps.

  • Inngest agent evals

    inngest/inngest-skills/skills/inngest-agent-evals Skill

    no license28★ repo

    Agent Skills for building with Inngest

  • Skill creator

    panaversity/agentfactory-business-plugins/.claude/skills/skill-creator Skill

    28★ repo

    Marketplace of domain-specific plugins for AI agents (Cowork, Claude Code, OpenClaw). Build autonomous business workflows for finance, banking, legal operations, and sales using modular agent skills and commands.

  • Skill creator

    a-tokyo/agent-skills/.agents/skills/skill-creator Skill

    15★ repo

    🧠 AI Agent skills for LLMs and AI Agents - Claude, Codex, Cursor etc.

  • Anti hallucination

    a-ariff/ariff-claude-plugins/plugins/anti-hallucination/skills/anti-hallucination Skill

    14★ repo

    65 plugins that turn Claude Code into an autonomous development team. 24 agents, 34 skills, 5 hooks. Includes 12-plugin anti-hallucination suite. One-line install.

  • Rails llm integration

    nagstler/rails-llm-integration Skill

    13★

    🔥🔥 A Claude Skill that teaches Claude Code how to write LLM features

  • Nasde benchmark calibration

    NoesisVision/nasde-toolkit/.claude/skills/nasde-benchmark-calibration Skill

    11★ repo

    CLI for benchmarks & evals of AI coding agents — on tasks you already understand, using your Claude / Codex / Gemini individual subscriptions or API keys.

  • Trtc eval

    Tencent-RTC/agent-skills/.claude/skills/trtc-eval Skill

    11★ repo

    TRTC AI Integration Assistant - helps developers integrate Tencent Real-Time Communication SDKs (Chat, Call, RTC Engine, Live, Conference) across Web, Android, iOS, Flutter, and Electron.

  • Eval surfer

    di37/EvalSurfer/skills/eval-surfer Skill

    11★ repo

    Skill-first, agent-native evaluation protocol for AI apps

  • Skill creator

    isomoes/skills/.agents/skills/skill-creator Skill

    10★ repo

    Custom skills for agents to support academic paper writing and thesis management

  • Skill creator

    rohit-yadav34/second-mind/.agents/skills/skill-creator Skill

    no license8★ repo

    Second Mind is a skill that scans your repository and conversation history to capture design decisions, open questions, and session context, storing everything as plain Markdown so project knowledge is portable, human-readable, and easy to resume.

  • Hallucination guard

    Momo2323-ui/claude-orchestra/skills/hallucination-guard Skill

    8★ repo

    An operating system for your Claude Code skills, agents & MCPs — organize everything into auto-routing orchestras.

  • Skill creator

    Victory-7291/project-scaffold-setup-skills/.agents/skills/skill-creator Skill

    7★ repo

    Reusable agent skills for scaffolding C++ and embedded projects

  • Benchmarking

    rusel95/ios-agent-skills/scripts/benchmarking Skill

    7★ repo

    Production-tested iOS Agent Skills for Claude Code, Codex, and 40+ AI coding tools. 8 enterprise-grade skills covering SwiftUI MVVM, UIKit MVVM, VIPER, TCA, Swift Concurrency, GCD, Testing, and Security Audit.

  • Eval driven dev

    yiouli/pixie-qa/skills/eval-driven-dev Skill

    7★ repo

    Agent skill for AI agent development

  • Skill creator

    Poorgramer-Zack/dart-expert-skills/.agents/skills/skill-creator Skill

    no license7★ repo

    A comprehensive library of modular Agent Skills for Flutter & Dart development

  • Testing with pitlane

    pitlane-ai/pitlane/skills/testing-with-pitlane Skill

    no license6★ repo

    Race a baseline vs a skill or MCP on real tasks. Hard checks show if it got better, faster, or cheaper. Numbers, not vibes.

  • Skill creator

    crowdin/skills/.agents/skills/skill-creator Skill

    5★ repo

    Agent Skills to help developers using AI agents with Crowdin

  • Skill creator plus

    yaniv-golan/skill-creator-plus/skill-creator-plus/skills/skill-creator-plus Skill

    4★ repo

    Based on Anthropic's skill-creator — with bug fixes, working Cowork support, and official best practices baked in. Claude skill for creating, testing, and improving other Claude skills.

  • Eval coach

    BayramAnnakov/eval-coach Skill

    4★

    Agent Skill for Evaluation-Driven Development (EDD) - guide AI evaluation strategies with the 50-40-10 rule

  • Benchmark agents

    build-with-dhiraj/ai-workflow-framework-portability-kit/Plugins/vercel-marketplace-source/.claude/skills/benchmark-agents Skill

    4★ repo

    Portable, self-contained snapshot of a complete Claude Code setup — 36 specialist agents, 134 skills, plugins, MCP servers & host tooling. Clone, claude login, run one script, restore the whole orchestration stack in ~20 min.

  • Agentic skill eval

    stempeck/agentfactory/.claude/skills/agentic-skill-eval Skill

    3★ repo

    Multi-agent orchestration CLI for Claude Code — declarative TOML workflows, autonomous agents, context-compression recovery, inter-agent mail.

  • Benchmark adder

    tmuskal/arc-agi-benchmarker/.claude/skills/benchmark-adder Skill

    no license3★ repo

    Prove you achieved AGI at home by testing your claude-code setup against arc-agi-3 benchmarks

  • Agent design doctrine

    urmzd/dotfiles/dot_agents/skills/agent-design-doctrine Skill

    3★ repo

    Cross-platform dotfiles managed by Chezmoi with Homebrew/apt and per-language version managers. One-command bootstrap for macOS and Linux with Neovim, Tmux, Zsh, and AI agent skills.

  • Skill creator

    drsh4dow/grug-brain-skill/.agents/skills/skill-creator Skill

    3★ repo

    A simplicity-first skill for reviewing architecture, refactors, APIs, and testing strategy without adding unnecessary complexity.

  • signet-eval

    jmcentire/signet-eval/io.github.jmcentire/signet-eval MCP server

    3★ repo

    Deterministic policy enforcement and MCP management for AI agent tool calls.

  • Skill creator

    harshitsinghbhandari/domain-expansion/.agents/skills/skill-creator Skill

    no license2★ repo

    Collection of Claude agent skills (code audits, LLM councils, PR review, resume tooling) installable individually via npx skills add.

  • Agent optimization

    abhishekgahlot2/agent-optimization/skills/agent-optimization Skill

    2★ repo

    Prompt, checklist, and Agent Skill for optimizing LLM agents without benchmark hacks.

  • Proofrag

    unshDee/proofrag/skills/proofrag Skill

    2★ repo

    Point your agent at your docs and your RAG app; get a golden test set + an LLM-as-judge & retrieval scorecard, in one command.

  • Eval

    croftspan/gigo/.claude/skills/eval Skill

    1★ repo

    義剛 GIGO. AI projects that get better every session, not worse.

  • Cdfi peer benchmark

    Jaypatel1511/cdfi-superpowers/skills/cdfi-peer-benchmark Skill

    1★ repo

    AI skills for NMTC eligibility, bank-CDFI peer benchmarking & HMDA analysis — grounded in audited PyPI tools, not hallucinated.

  • Skill creator

    Modellix/modellix-skill/.agents/skills/skill-creator Skill

    1★ repo

    An Agent Skill of Modellix documentation and capabilities reference.

  • Cost quality frontier

    varunk130/AI-Eval-Skills/skills/cost-quality-frontier Skill

    1★ repo

    Curated AI agent evaluation skills from Microsoft's Eval Guide — plan, generate, run, and interpret eval suites for Copilot Studio agents

  • Ai workflow

    vindm/dotclaude/deferred/skills/ai-workflow Skill

    1★ repo

    AI dev infrastructure framework for Claude Code. /dotclaude:bootstrap authors CLAUDE.md + docs/ + .claude/ tuned to your project. Per-domain: design (showpiece), coding, planning, testing, data, ai-workflow.

  • Skill creator

    arcjet/skills/.agents/skills/skill-creator Skill

    1★ repo

    Skills to make your AI coding agent an Arcjet security expert.

  • Skill creator

    morzecrew/agent-skills/.agents/skills/skill-creator Skill

    1★ repo

    Collection of agent skills

  • Setup skills evals

    ahnafyy/skills-evals/skills/setup-skills-evals Skill

    1★ repo

    Catch when your agent skills, instructions, and rules stop triggering — or stop working.

  • Skill creator anthropics 20260226

    MatrixFounder/Universal-skills/archive/skill-creator-anthropics-20260226 Skill

    1★ repo

    Collection of high-leverage "Meta-Skills" designed to upgrade AI Agents from simple chat bots to autonomous engineers

  • Skill creator

    jtmthf/skills/.agents/skills/skill-creator Skill

    no license1★ repo

    Personal Agent Skills

  • Skill creator

    Metaverse-Cloud/EngageLab-Skills/.agents/skills/skill-creator Skill

    1★ repo

    EngageLab-Skills

  • Repo evals

    zinan92/repo-evals Skill

    1★

    Claim-first repo 评测框架。in target repo + claim map → out bilingual verdict dossier + all-evals dashboard

  • Skill creator

    krkn-s/ai-workspace/.pi/skills/skill-creator Skill

    1★ repo

    KRKN's Personal AI workspace • Skills / Prompts / ...

  • Skill creator

    eagerworks/skills/.agents/skills/skill-creator Skill

    1★ repo

    A collection of portable agent skills by Eagerworks — install via the skills.sh CLI for Claude Code, Cursor, Copilot, Codex, Amp, and more.

  • Candidacy checker eval

    0SxD/candidacy-checker-eval Skill

    1★

    Generic two-layer candidacy evaluation framework: Socratic profile-discovery loop with dialectic confidence saturation, plus a flat boolean rubric.

  • Vercel eve framework

    Xtqm/skills/skills/vercel-eve-framework Skill

    1★ repo

    I'll be adding my personal agent skills into this repo.

  • Skill creator

    udohjeremiah/skills/.agents/skills/skill-creator Skill

    1★ repo

    Reusable Agent Skills for AI coding agents, focused on software engineering workflows.

  • Add skill

    Paldom/github-skills/.claude/skills/add-skill Skill

    1★ repo

    Agent Skills for professional GitHub repos - brief, structured, SEO-friendly READMEs that convert visitors into users, plus complete OSS scaffolding: community health files, templates, and discoverability best practices.

  • Add skill

    Paldom/icon-designer-skills/.claude/skills/add-skill Skill

    1★ repo

    Agent Skills that design minimalist app and OSS package icons from a text brief or project context - symmetric logos on dark grey, Apple-style rounded-rectangle backgrounds.

  • Add skill

    Paldom/skillskit/skills/add-skill Skill

    1★ repo

    From context to installable agent skills - research packs in, validated skills.sh-ready skills out. Scaffolding, eval-first authoring, validation, and deployment included.

  • Add skill

    Paldom/python-skills/.claude/skills/add-skill Skill

    1★ repo

    Agent Skills and hooks for maintaining high-quality open-source Python packages - an agentic engineering setup covering linting, testing, packaging, releases, and CI quality gates.

  • Add skill

    Paldom/node-skills/.claude/skills/add-skill Skill

    1★ repo

    Agent Skills for maintaining high-quality open-source Node.js, TypeScript, Next.js, and React apps and packages - linting, type safety, testing, packaging and releases, and CI quality gates.

  • Benchmark price

    nathan8823/fairbill/.claude/skills/benchmark-price Skill

    1★ repo

    Turn your AI agent into a medical-bill negotiation advocate — an open-source playbook of skills, letters, and verified patient rights. Pay what's fair, nothing more.

  • Evals mcp server

    cyanheads/evals-mcp-server/io.github.cyanheads/evals-mcp-server MCP server

    1★ repo

    Author verifiable eval records through a draft→review→revise→submit loop with enforced graders.

  • Contrastive authority eval

    jpoindexter/reward-seeking-safety-skills/skills/contrastive-authority-eval Skill

    0★ repo

    Agent skills and eval utilities for reward-seeking, grader-targeting, and oversight-dependent behavior.

All 944 in the catalog