skill-creator

Skill Creator

Create one focused, reusable Skill package. Skills are application-managed packages, not Artifacts. Use the JavaScript control-plane REPL and the native host.skills composer for lifecycle operations.

Native composer

await host.skills.list()
await host.skills.read(name)
await host.skills.read(name, path)
await host.skills.validate(name)
await host.skills.edit(name, path, content)
await host.skills.edit(name, path, replacement, oldString)
await host.skills.publish(name)
await host.skills.publish(name, true)
await host.skills.delete(stableId)

Without oldString, edit creates a file and fails if it exists. With oldString, the old text must occur exactly once. Never silently overwrite an existing draft file. publish promotes the complete draft into Personal Skills. delete is privileged and always uses app approval. When a published Skill and its draft coexist, delete only by the exact draft-<name> or personal-<name> id returned from list(); never guess from the shared display name.

Choose the current stage

Infer where the user is in the workflow and start there:

1. Capture intent and examples.
2. Draft or revise the Skill.
3. Review the package with the user.
4. Optionally evaluate realistic prompts.
5. Improve from evidence and repeat.
6. Publish only after the user accepts the draft.

Do not force evaluation. Objectively verifiable workflows benefit from test cases; subjective writing or exploratory Skills may be better reviewed directly in conversation.

Capture intent

Extract what is already known from the conversation before asking questions. Confirm only gaps that materially change behavior:

• What should the Skill enable an agent to do?
• When should it trigger, including near-miss cases where it should not?
• What inputs and output formats matter?
• What counts as success, and what failures need explicit handling?
• Are scripts, references, assets, connectors, or example files required?
• Does the user want test cases now?