# Adopts a published skill set without giving up control of the project

> Competency customizing-agents/adopts-a-skill-set

Page: https://schubergphilis.github.io/ai-training/competencies/customizing-agents/adopts-a-skill-set/
License: CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/

*Customizing agents* · competency `customizing-agents/adopts-a-skill-set`

**Taught in:** the [Customizing agents](https://schubergphilis.github.io/ai-training/customizing-agents/) course

**Draws on:** [Agent skills](https://schubergphilis.github.io/ai-training/topics/customizing-agents/skills/), [Project instructions](https://schubergphilis.github.io/ai-training/topics/customizing-agents/instructions/), [Hooks, permissions, settings](https://schubergphilis.github.io/ai-training/topics/customizing-agents/hooks-permissions/)

## Learning objectives

### Vets a published skill set before installing it, as they would a dependency (base)

| Claim | Why | Example |
| ----- | --- | ------- |
| The learner checks the license of the set and of each skill they plan to use. | One collection can mix licenses, with open-source skills next to skills that are only source-available (Agent Skills (the anthropics/skills repository), GitHub). | The learner finds that the document skills in a collection are source-available and uses only the Apache-licensed ones in the team's repository. |
| The learner looks at who maintains the set, how recently it changed and how its changes are released. | A skill set is code that runs in every session, and one that nobody maintains drifts away from the agent it was written for. | The learner sees regular releases with a changelog and an open issue tracker, and notes the version they reviewed. |
| The learner reads every skill, script and hook in the set before installing it. | A plugin's hooks and scripts run in the learner's sessions with the session's permissions (Plugins overview, Claude Code documentation). | One skill runs a setup script that writes to the home directory, so the learner reads the script before deciding. |
| The learner compares the set with the alternatives, including writing the few skills they need. | A set of thirty skills brings thirty descriptions and thirty things to keep up to date, and the team may need three of them. | The learner writes their own review skill from the two ideas they liked, and does not install the whole collection. |

Served by: [Adopting a published skill set](https://schubergphilis.github.io/ai-training/customizing-agents/adopting-a-skill-set/)

### Adopts a few skills at a time and checks each one's effect and context cost (base)

| Claim | Why | Example |
| ----- | --- | ------- |
| The learner turns on one or two skills at a time and uses them on real work before adding more. | With one change at a time, a better or worse result can be traced to the skill that caused it. | The learner uses the planning skill for a week of tickets before trying the review skill from the same set. |
| The learner checks how much context the installed skills add to every session. | The description of every model-invoked skill loads in every session, whether or not the skill is used (Extend Claude with skills, Claude Code documentation). | The context breakdown shows the skill descriptions growing after the install, so the learner makes two rarely used skills user-invoked. |
| The learner removes a skill that did not change the results. | A skill with no effect still costs context and still has to be kept up to date. | After two weeks the commit-message skill produced the same messages as before, so the learner uninstalls it. |

Served by: [Adopting a published skill set](https://schubergphilis.github.io/ai-training/customizing-agents/adopting-a-skill-set/)

### Keeps the project's own instructions, hooks and permissions in charge when a skill set disagrees (base)

| Claim | Why | Example |
| ----- | --- | ------- |
| When a skill contradicts the project's instructions, the learner changes or drops the skill and leaves the instructions as they are. | When two instructions conflict, the agent may follow either one, so the conflict must be removed (How Claude remembers your project, Claude Code documentation). | A skill says to commit after every change, the project says to commit once per task, and the learner removes that step from their copy of the skill. |
| The learner keeps enforcement in the project's hooks and permissions, and no skill widens them. | A skill is text the agent reads, and only hooks and permission rules enforce anything (Best practices for Claude Code, Claude Code documentation). | A skill asks for permission to run any shell command, and the learner keeps the narrow allowlist and adds only the one command the skill needs. |
| The learner pins the version of the set and reviews each update before taking it. | An update changes what runs in every session, and a pinned version changes only when someone decides it should (Install and manage plugins, Claude Code documentation). | The team turns off automatic updates for the marketplace and reads the changes of each new release before moving the pin. |

Served by: [Keeping the project in charge](https://schubergphilis.github.io/ai-training/customizing-agents/project-stays-in-charge/)

## References

- Agent Skills (the anthropics/skills repository), GitHub, https://github.com/anthropics/skills
- Plugins overview, Claude Code documentation, https://code.claude.com/docs/en/plugins
- Extend Claude with skills, Claude Code documentation, https://code.claude.com/docs/en/skills
- How Claude remembers your project, Claude Code documentation, https://code.claude.com/docs/en/memory
- Best practices for Claude Code, Claude Code documentation, https://code.claude.com/docs/en/best-practices
- Install and manage plugins, Claude Code documentation, https://code.claude.com/docs/en/plugins/install