Support project-level skills discovery from repository workspace
Open 💬 2 comments Opened May 9, 2026 by urolzeen
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What variant of Codex are you using?
App, IDE Extension, CLI
What feature would you like to see?
Codex currently discovers skills mainly from user-level ~/.codex/skills, but project-specific skills are not automatically loaded from the repository workspace.
This makes team workflows fragile:
- Skills are easy to miss after reinstalling the OS or switching machines.
- Project-specific AI workflows cannot travel with the repo.
- Different developers may have different local skills installed.
- AGENTS.md and project rules are workspace-aware, but skills are not, which feels inconsistent.
Expected behavior:
- Codex should automatically discover workspace-level skills from a conventional repo path such as:
- <repo>/.codex/skills/<skill-name>/SKILL.md
- or <repo>/.agent/skills/<skill-name>/SKILL.md
- The available skills list should show where each skill comes from:
- system
- user
- workspace/project
- plugin
Security:
- Since project-level skills come from the repository, Codex could require a one-time workspace trust confirmation before loading them, similar to VS Code Workspace Trust.
- If two skills have the same name, Codex should define a clear precedence rule or report the conflict explicitly.
Why this matters:
This would make Codex much easier to use in team repositories and portable development environments. Project-specific AI workflows could be versioned together with the codebase instead of requiring every developer to manually copy skills into ~/.codex/skills.
Additional information
_No response_
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Hi @urolzeen — your points about project portability, inconsistent local Skill installations, source visibility, and duplicate conflicts are exactly the problems I’m researching.
I’m building Bitcase, a free public Alpha that collects external Skills, records their sources, checks SKILL.md files, detects duplicate content, and assembles a project-specific Skill Stack that can be copied into Codex.
Bitcase does not yet write Skills into a repository-level .codex/skills directory or implement workspace trust. One question I’m trying to answer is whether a project-specific Stack and source manifest are useful first steps, or whether repository-level installation must exist before the product creates real value.
Would you be willing to test it with one real project and tell me where that model fails?
https://bitcase-skill-library.tomcjq070814.chatgpt.site/?utm_source=github&utm_campaign=alpha_research&utm_content=urolzeen_21907
No account or private repository is required.