Skip to content

Add extra Skills

Prepared, unreleased authoring contract. This public page documents a reviewed future authoring workflow. Future authoring commands described here are not available in current releases; they do not claim that plugin-kit-ai@2 is available. Existing installer examples are identified separately. Public activation remains pending.

Use the future skills init job to add one Skill to an existing standard package. It works for Skill, MCP, or hybrid packages. Select the package root containing plugin.json, not the skills/ directory or an individual SKILL.md file.

Name the new Skill and package

Assume ./review-helper is the package from the standalone Skill journey and skills/release-notes/ does not exist.

Future installer-hosted spelling:

bash
agentplugins author skills init release-notes ./review-helper \
  --description 'Use when drafting release notes from reviewed changes'

Equivalent future authoring executable spelling; choose one:

bash
plugin-kit-ai skills init release-notes ./review-helper \
  --description 'Use when drafting release notes from reviewed changes'

The positional Skill name is required, as is an explicit description of 1–1024 characters. Names are not normalized. Use an exact lowercase portable name with hyphens, and choose a new destination instead of expecting an overwrite.

Review the additional component

text
review-helper/
  plugin.json
  skills/
    review-docs/
      SKILL.md
    release-notes/
      SKILL.md

Edit skills/release-notes/SKILL.md to explain its trigger, method, and expected output. Keep its name and description accurate. For example, instruct it to separate user-visible changes from maintenance changes and to cite the reviewed source for each release claim.

Scripts, references, and assets can support instructions, but their presence is not proof of execution or permission to run them. The allowed-tools field is not independent authorization evidence. Review supporting material with the same care as the instruction text.

Validate the package and immediate Skills

bash
agentplugins author skills validate ./review-helper
plugin-kit-ai skills validate ./review-helper

These are equivalent checks; one entrypoint is sufficient. The command reports package-wide readiness and isolated immediate Skills. An error elsewhere in the package can therefore affect the overall result even when the new Skill's own findings are clean.

Omitting the package path uses only the current directory, with no ancestor search. Explicit roots make scripts and troubleshooting clearer. Running from inside skills/release-notes/ is not a substitute for selecting the package root.

Handle a failure without overwriting work

Read committed and the accompanying error guidance if creation fails. A Skill can be committed before resulting package checks fail or remain incomplete. Inspect the actual files and the report before retrying; a second init against an existing destination will not overwrite it.

Correct the reported input or package issue and repeat validation. If the name is wrong, review the intended source change explicitly rather than adding a force flag. No external Skills install, update, remove, or generate lifecycle is part of this v2 authoring journey.

Continue with static package checks and the handoff.

Public docs for plugin authors and integrators.