Tightening the codebase-onboarding skill
A teammate just shared a personal skill called codebase-onboarding to help new hires get oriented in a service repo, and it's already installed at ~/.claude/skills/codebase-onboarding/SKILL.md.
There are two problems with it that waste context every time the skill loads:
- It leaves Claude with full edit and write permission (the skill is supposed to be read-only. New hires shouldn't get an agent that modifies code while they're just trying to look around).
- The body is huge because the entire architecture reference was pasted directly into
SKILL.mdinstead of split into a separate file.
In this exercise you'll fix both: add an allowed-tools allowlist that restricts the skill to read-only discovery, and apply progressive disclosure by moving the architecture reference into its own file under references/
This exercise is part of the course
Introduction to Agent Skills
Exercise instructions
- Start Claude Code with
claude. - Ask Claude to read
~/.claude/skills/codebase-onboarding/SKILL.mdand identify any design issues. - Ask Claude to add an
allowed-toolsfield to the frontmatter restricting the skill to read-only discovery tools. - Ask Claude to apply progressive disclosure by moving the architecture reference into
references/architecture.mdand replacing it with a link. - Type
/exitto exit Claude Code.
Hands-on interactive exercise
Turn theory into action with one of our interactive exercises
Start Exercise