SkillJetty — Agent Skills Studio

Install the stable release of DavidFarris.skill-builder for the established workbench, or choose Install Pre-Release Version in VS Code to try the public preview. Both channels use the same Marketplace identity and SkillJetty: Open Workbench command; they are alternate versions in the selected VS Code profile. The separate local UAT extension remains a distinct installation.

The 0.11.4 public preview identifies itself inside the workbench and stores its drafts, history, packages and repository connection in .skilljetty-prerelease beneath the selected work folder. It does not automatically import or migrate stable or local-beta data. Absolute or escaping inherited skills/export directory settings fall back to channel-local defaults. Import an existing skill explicitly when you want a separate preview copy. Preview test results do not establish native accessibility or human UAT acceptance.

Agent Skills Studio — Build it. Improve it. Share it.

Create, edit, check and publish Agent Skills from one guided VS Code workbench.

Focus on what your skill should do. SkillJetty reads its declared identity, manages release numbers and prepares factual file-change notes. Save an installable package locally without a repository, or connect one when you want to share for review.

The preview acknowledges clicks with an activity indicator and explains what is happening. Expensive local checks run in the background; routine saves no longer reopen every saved release. See performance measurements and feedback behavior.

Open the workbench

Click the SkillJetty icon on VS Code's left activity bar, or run SkillJetty: Open Workbench from the Command Palette. On the preview channel, confirm the Public pre-release 0.11.4 banner before editing. Switching channels changes the installed version in that profile; it does not migrate your work data. Use a separate VS Code profile for an isolated evaluation alongside a protected stable setup.

Choose how to start, then follow the highlighted Next step action:

1. Start: choose Create a new skill, Open a .skill file, or Open a skill folder. You can also resume saved work. A repository connection is needed only when you are ready to publish.
2. Edit: write the skill's instructions and edit its supporting files in the workbench. Add text files or copy existing assets. Changes save locally.
3. Check: see problems with links to the files that need attention. Fix them, then check again. Checking does not upload anything, and a passed structural check does not establish model behavior.
4. Submit: review the detected file changes and managed release number. Add optional context, then save a package on this computer. To share through GitHub, connect a repository and choose Upload & create PR. You can optionally choose a reviewer; requesting review notifies that person. A manual release-number override is available under Advanced.
5. Done: open the pull request or the source, .skill and changelog links. Use Check status to refresh review progress. Choose Work on another skill when finished.

No terminal commands or Git branch names are needed for normal use. The extension detects the repository's actual default branch. Git credentials and repository access still need to work on your machine.

What is saved in the repository?

Skills/
  meeting-notes/
    1.0.0/
      source/meeting-notes/
        SKILL.md
        references/...
        assets/...
        scripts/...
        evals/...
        agents/...
      meeting-notes-1.0.0.skill
      CHANGELOG.md
      release.json
      original.skill           # only for an originally imported package

Only existing supporting folders are included. Complex nested files and binary assets are preserved. First versions need changelogs too. Past versions are immutable; publish a new version for changes.

For GitHub.com, submission uploads a review branch and creates a pull request, reusing an existing matching PR when retried. Review and merge follow your repository's rules. Approval alone does not put the skill on the default branch; the workbench never merges automatically. It does not upload the skill into an AI application or publish this extension to the Marketplace.

Use Check PRs from Start to refresh saved submissions, including uploads made with older SkillJetty versions. If an older upload has no PR, choose Create PR. Refreshing status never creates a PR or requests a reviewer. Status shows awaiting review, changes requested, approved, merged, or closed without merge. If GitHub cannot be reached, the workbench shows the error and last verified status rather than guessing.

This review flow supports GitHub.com. Other Git destinations retain branch uploads; their review process must be completed in the hosting service. Additional repository services are a future discussion.

Portable skills and application checks

The default is the open Agent Skills format: SKILL.md plus optional supporting files. You can select Portable, Claude, ChatGPT/Codex, Gemini CLI or GitHub Copilot for compatibility guidance. Application-specific files such as agents/openai.yaml are preserved. Compatibility checks do not promise identical behavior across applications, install the skill, or run a model.

Your work stays recoverable

• Failed checks, canceled file pickers and failed uploads retain your draft.
• A completed submission removes an unchanged draft from active work and archives it recoverably. A partial submission retains recovery actions; retrying PR creation does not upload the version again.
• Newer unpublished changes remain active.
• Resume saved drafts or choose a library version to edit its next version.
• If another editor changes a file, conflicting workbench edits are saved in local recovery records rather than overwriting the newer file.
• The UI uses text and icons alongside color. There are no flashing prompts; reduced-motion and high-contrast preferences are respected.

See the short workflow and recovery guide.

Requirements and development

VS Code 1.90+. Publishing requires Git, a configured commit identity, and repository access. GitHub.com PR creation, reviewer selection and status checks use VS Code's GitHub sign-in. Git push access and API access are separate: SkillJetty checks API access before a fresh GitHub submission. Use Connect GitHub account or Switch GitHub account if the signed-in account cannot access the destination. Your repository or organization may require additional authorization.

Repository discovery and creation still use your signed-in GitHub CLI account; entering an existing repository URL does not require that discovery step. The new PR tools do not require the CLI. No model credentials, paid service, or global validator installation is required. SkillJetty does not store your GitHub token in project files.

npm install
npm test
npm run package

Older authoring commands remain internal for compatibility. The guided workbench is the normal entry point. Local preview builds may be newer than the Marketplace listing.

MIT License. See LICENSE.