skillrank: Find and Evaluate AI-Agent Skills

Overview

SkillRank helps developers find and manage SKILL.md packages for coding agents, then assess their practical effect instead of relying on popularity or install counts. Its local evaluation harness runs paired trials with and without a skill on the developer's agent and reports changes in success, token use, turns, time, and cost.

The CLI can be used without an account for search, installation, recommendations, and local evaluation. Contributions such as publishing skills and reviews require login. SkillRank also offers MCP integration for Claude Code and Codex, and can run a local registry with a seed catalog.

Key Features

• Search and inspect skills, with filters for stack, agent, and category.
• Install skills with content-hash verification, lockfile tracking, and drift reporting.
• Recommend skills based on detected repository stack.
• Run paired skill-versus-no-skill evaluations on the user's agent and report per-task metric deltas.
• Keep evaluation results separated by trust tier: official, community-reported, and self-reported.
• Register SkillRank tools with Claude Code and Codex through an MCP server.
• Serve a local registry for search, recommendations, and installation without relying on a hosted registry.
• Publish skills and contribute ratings, reviews, and evaluation results with an account.

Use Cases

• A coding-agent user wants to check whether a skill changes task effort or outcomes before adding it to a project.
• A developer wants to discover skills suited to a repository's framework or toolchain.
• A team wants hash-verified skill installation and a record of installed skills in its repositories.
• A skill author wants to publish a package and contribute evaluation results or reviews to the registry.
• A user of Claude Code or Codex wants to search for and manage skills through native agent tools.

Project Facts

• Language: Rust
• License: MIT
• Stars: 0
• Forks: 0
• Topics: none
• Archived: no

Getting Started

Install with the project installer:

curl -fsSL skillrank.dev | sh

Then try skillrank search playwright or skillrank recommend. See the README for evaluation setup, MCP integration, configuration, and build instructions.

Alternatives

• clawhub: ClawHub provides a shared registry and install workflows for OpenClaw skills, while SkillRank focuses on discovering, installing, and evaluating skills across coding-agent runs.
• SkillOpt: SkillOpt iteratively improves skill documents using scored task trajectories; SkillRank measures the effect of enabling a skill against a control run.
• SkillSpector: SkillSpector checks skills for security risks before installation, while SkillRank evaluates their impact on task success, tokens, time, and cost.

Considerations

• Evaluation results depend on the agent, suite, and trial count. The README cautions that results from fewer than five trials per arm are directional, not significant.
• Evaluations run on the user's agent and use task suites with pinned fixture repositories. Cost may be unavailable when an agent or trial does not report it.
• Search and install use a registry over HTTP. A local registry is available; publishing and other contributions require an account.
• The repository's input metadata lists MIT, while the README distinguishes the CLI and core library, which are MIT, from the hosted registry and website, which use Elastic License 2.0. Check component licenses if self-hosting or redistributing those services.
• The repository lists 0 stars and 4 open issues, so public adoption and issue activity are limited indicators of maturity.