21 October 2026 · Claude Code skill-loading check
Installed a frontend skill? Check whether it ran.
Before comparing outputs: confirm that the target agent loaded the skill. In Claude Code, an installed skill's description can be available while its full instructions remain unloaded.
Claude Code's skill documentation says the full SKILL.md body loads when the skill is invoked. Claude may choose a skill when relevant; you can call a user-invocable skill directly with /skill-name. This is a Claude Code rule, not a rule for every agent.
A short check for Claude Code
| Check | Why it matters |
|---|---|
| Ask “What skills are available?” and look for the intended skill. | If it is missing, a trial cannot test that skill. |
Use /skill-name when the skill is user-invocable, or confirm a skill call in the session. | Availability alone does not establish that its body loaded. |
| If automatic use fails, inspect the skill description and frontmatter. | Claude Code's troubleshooting guide names both as possible causes. |
Then hold the model, project, task brief, and acceptance checks fixed for a no-skill comparison. Record the source revision and instructions that loaded. Check whether the output reuses your components and tokens, handles required states, and works in the browser. A skill call proves loading, not better design.
Evidence status: official documentation inspected; no pool-run installation, invocation, or frontend outcome test.