Announcing skills on Tessl: the package manager for agent skills

Today, we’re announcing agent skills on Tessl: a developer-grade package manager for skills, tools to evaluate their quality, a registry full of evaluated skills, and a platform to manage the full lifecycle of skills in your organization.

From today, you can:

• Find evaluated skills on the Tessl Registry
• Install Tessl and search for a skill in the CLI tessl skill search or install the skill directly with with Tessl install e.g. tessl i github:softaworks/agent-toolkit
• Publish and evaluate your own skills by using tessl skill publish

Read on to learn more about evaluations, the launch and professional development of skills, or dive into the docs to get more hands on!

Professional dev tools for skills

Agent skills have made AI coding agents more powerful than ever, adapting to every environment and task. But as the usefulness of skills increases, so does the cost of managing them.

Today, most teams still treat skills as static artifacts: markdown files, created or copied from repo to repo. This approach offers a strong initial boost, but quickly creates debt:

• Skills are duplicated and updates never roll out.
• Poor quality skills go unseen, misguiding agents instead of helping.
• Skill knowledge grows stale, not keeping up with the systems the practices they describe.

Over time, skills that once helped agents become a liability.

Skills are more than markdown: they’re the next unit of software, and they need professional developer tooling.

Skills need to be evaluated

As teams rely more on skills, it quickly becomes clear that the impact of the same skill on different agents and model varies, and that not all context helps.

Some skills are too long, overwhelming the context window and hindering task completion. Other skills lack critical information, eliminating their impact. And some skills are simply phrased poorly, which some models can handle and others can’t. Since most skills are generated by agents (with a skill!), mistakes are also very common.

Without a way to evaluate skills, teams have no clear way to understand how good a skill actually is, or if it degraded over time.

Quoting a DevOps true-ism: you can’t optimize what you can’t measure. Evaluations are the foundation for having quality skills — not by assuming more context helps, but by measuring how skills influence agent behavior and improving them deliberately over time.

How to evaluate your skills

Tessl currently evaluates skills in two ways: review evals and task evals.

Review evaluations test the structure of the skill by reviewing it against Anthropic’s best practices. These practices show the best way to get models and agents to understand your instructions, and tuning to them truly helps. Here are some examples of review eval results:

The Tessl Registry already contains review evals for over 2,000 skills, and you can request an evaluation for any public skill here!

Task evaluations test the skill’s instructions by running agents with real tasks, with and without the skill, and assessing the results. These powerful evaluations demonstrate the efficacy of the skill on the agents and models you’re actually using, allowing you to optimize skills and remove poorly performing ones.

Task evals depend on accurate eval scenarios and judgement metrics, which are typically tuned by the owner and users of the skill. Tessl facilitates this process, and we’re in the process of generating eval data for a large set of public skills - stay tuned! You can get a taste of what they might look like by seeing the evals we’ve generated on library docs.

A package manager for agent skills

Skills are designed to be reused, but currently lack a distribution system. To consume a skill, developers commit a copy to their own repos, creating duplicates with no means of tracking or updating. It’s even common to see duplicates in the same repo to support different agents!

The software world already has a solution for software distributions - package managers. Package managers made software scalable by making dependencies explicit, versioned, and reusable. Agent skills need the same treatment.

Tessl provides exactly that - a package manager for agent skills and context. Skills can be versioned so updates can clearly be seen. Skill installation is captured in a manifest file, clarifying which skills are installed and helping upgrades. And the Tessl Registry helps you discover skills, choose based on their evaluated quality, and consume them over time.

To install skills with Tessl, start by installing tessl by running npm i -g @tessl/cli. You can then use Tessl to install skills from the registry or any GitHub URL using the right ID (e.g. tessl i github:vercel/ai) or direct Tessl ID (tessl i tessl-labs/example). You’ll notice a tessl.json file records the installation, allowing future tessl install and much more.

Want to publish your own skills to the registry to get them reviewed and distributed? Register for a free account, then use tessl skill publish and point it to your skill folder.

Not sure what to install? Discover evaluated skills in the Tessl Registry, or check out our docs to learn more.

Lifecycle management for agent skills

Once skills are treated as software, the requirements for owning them become clear.

Skills need tests to let them evolve over time without breaking behavior. Teams need to know what versions are installed, what changed, and how updates affect outcomes. They need evaluations and visibility so behavior changes are observable, not implicit. And when skills are shared and reused, they need dependency-style management so adoption and upgrades are intentional.

Tessl provides the full lifecycle for agent skills and context: build, evaluate, distribute, and optimize. Today’s launch focuses on Tessl’s skills package manager and evaluations, but those are just a part of the full solution. If you’d like to explore the full Tessl Agent Enablement platform for your organization, contact us.

Get started for free

Discover evaluated skills in the Tessl Registry, or install and test any skill from GitHub.

npm i -g @tessl/cli && tessl skill search

Need help? Read the skills quick start guide or speak to our team in Discord.