OpenAI Discontinues Codex Skills Collection, Shifts to Plugins

OpenAI has discontinued the skills collection it offered for Codex. It is directing users to the plugins collection as the migration destination. This was revealed in information published on 2026-09-09. Developers will now need to follow the procedure for creating plugins. The mechanism that packaged repetitive work into reusable components has reached a turning point.

Reporting on openai via GitHub Trending describes the situation as follows.

Important This repository is deprecated. For current Codex skill and plugin examples, use the OpenAI Plugins repository.

Development workflows that relied on the old collection will need to be reviewed. Skills distribution is being consolidated into plugins.

Discontinuation of the Developer Skills

Collection and Migration Destination

The legacy skills collection has been discontinued and will no longer be updated. Current skill and plugin examples are provided in the OpenAI Plugins collection. The same applies when adding custom skills to Codex; it is not the same collection. Refer to the Build plugins guide, the procedure for creating plugins. That procedure also includes how to create skill-only plugins.

The discontinuation covers the entire collection containing skills for Codex. No fixes or additions to individual skills will be accepted in the old collection. Users need to switch where they look for examples. Updating references in development documents and internal procedures will also be an issue. Consolidating the migration destination is expected to eliminate fragmented information.

On related operational aspects, management of automated processes as shown in OpenAI Agents Abused Wiki in 18,000-Post Information Sharing Incident is also a point of discussion. This is a stage where both the power of skills and control over them are required.

Definition and Mechanics of Agent Skills

An agent skill is a unit that bundles instructions, scripts, and resources. An AI agent detects and uses it to perform a specific task. Its distinctive feature is that once written, it can be used in multiple situations. Codex uses skills as a means of performing repeatable work. It serves to package and share the capabilities of individuals and teams.

A skill is not merely a fragment of instructional text. It consists of execution procedures, supporting scripts, and reference materials. It functions as a reusable component focused on a specific task. It aims to lock in quality and reduce variation in procedures. It is suited to automating routine processes in development support.

The diversification of foundation models shown in OpenAI Announces GPT-5.6 With Three Models: Sol, Terra, and Luna is also expanding the use cases for skills. The combination of models and skills determines work efficiency.

Skill Installation Process in Codex

In the latest version of Codex, built-in skills are installed automatically. They become effective without any individual action by users. Curated skills and experimental skills require manual installation. A procedure using Codex’s built-in installation helper was described.

Curated skills can be installed by specifying their names. The default reference is skills/.curated. Pass the skill name to the installation helper and run it. Experimental skills require specifying their storage location. As an example, create-plan in the .experimental location was cited.

URL-based specification pointing to a directory location on GitHub is also supported. After installation, Codex must be restarted. The mechanism loads new skills upon restart. Thorough restarts are important as an operational procedure. Users need to understand that changes are not applied automatically.

Open Standard and Integration Policy

Toward Plugins

Agent Skills are being developed as an open standard. Details of the standard are provided in a separately published document. Skill usage in Codex is also explained. How to create custom skills is likewise documented. The direction points toward improving skill portability.

This discontinuation is not a discontinuation of the standard itself. It is a measure to consolidate the distribution collection on the plugins side. A form called skill-only plugins has been prepared. They can be registered as plugins containing only skills. Existing skill assets can be carried over in plugin form.

Conversion to plugins clarifies the unit of management. Permissions, sharing scope, and version control can be handled per plugin. Sharing and distribution within teams and organizations become easier to organize. It is seen as easier to maintain operational control than with standalone skill distribution.

Impact on Development Workflows and

Practical Response

The practical impact centers on changes to references. Documents that referenced the old collection will need to be rewritten. The source for examples will switch to the plugins collection. Procedures for creating new skills will also be unified into the plugin procedure. Updating developer training materials will be the immediate task.

Skills subject to automatic installation are less affected. Skills included in the system can continue to be used. Curated and experimental skills installed manually need to be checked. Work is needed to verify whether they are offered in the plugins collection. Listing dependencies is the starting point for migration.

Legal confirmation is also essential. Terms of use for individual skills are shown in the LICENSE.txt within each skill. Conditions apply per skill, not for the collection as a whole. The procedure for verifying rights will remain even after conversion to plugins. As shown in Seattle Times and Others Sue OpenAI and Microsoft, rights clearance for training materials is becoming stricter. Managing the provenance of materials included in skills is becoming more important.

Editorial Opinion

In the short term, existing users will face migration work. Document locations are changing, and installation procedures need to be confirmed. Moves to reconfirm the distinction between automatically installed and manually installed targets are expected to spread. In the field, auditing dependencies and updating documents will be the immediate challenges.

In the long term, the distribution channel for skills is expected to be unified into plugins. The shift from a standalone collection to the plugin format will improve reusability. If alignment with the open standard is maintained, interoperability with other products will also advance. As a result, packaging repetitive work into components could take root in development culture.

The remaining issues are post-migration compatibility and where operational responsibility lies. Will procedures dependent on the old collection continue to work as-is. How will permission management and review change with conversion to plugins. The question of who guarantees skill quality is also being raised.

References

• “openai /

  skills", by **openai** — GitHub Trending, 2026-09-09 (ARR)

• Source URL: https://github.com/openai/skills