[BUG] Plugin skills not directly loaded when invoked via slash command — spawns unnecessary Explore agents

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When invoking an installed plugin skill via its namespaced slash command (e.g., /playwright-tooling:playwright-automate), Claude Code does not directly load the skill. Instead, it spawns Explore agents to search the filesystem for the skill and related scripts — even though the plugin is installed and the skill path is already known.
This wastes significant tokens (~46k in my case) and delays execution for something that should be an immediate, direct load.

Steps to Reproduce

Install a plugin from the marketplace (in my case, playwright-tooling)
Verify it shows as installed via /plugin list
Invoke a skill via its namespaced command: /playwright-tooling:playwright-automate <arguments>
Observe that Claude spawns Explore agents to search for the skill rather than loading it directly

What Should Happen?

When a plugin skill is invoked by its exact namespaced command, Claude should:

Resolve the skill directly from the installed plugin's directory structure
Load the SKILL.md / COMMAND.md content
Execute the instructions immediately

No Explore agents should be needed — the skill's location is deterministic from the plugin installation.

Error Messages/Logs

❯ /playwright-tooling:playwright-automate For test teshiba, document a tdap immunization and cat hair allergy
● Running 2 Explore agents… (ctrl+o to expand)
   ├─ Find playwright-automate skill · 11 tool uses · 23.8k tokens
   │  ⎿  Searching for 11 patterns…
   └─ Find playwright-cli scripts · 8 tool uses · 22.4k tokens
      ⎿  Searching for 8 patterns…

Steps to Reproduce

Install a plugin from the marketplace (in my case, playwright-tooling)
Verify it shows as installed via /plugin list
Invoke a skill via its namespaced command: /playwright-tooling:playwright-automate <arguments>
Observe that Claude spawns Explore agents to search for the skill rather than loading it directly

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.79

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

This appears to be part of a broader set of issues with plugin skill visibility and invocation:

user-invocable: true on skills does not surface them in the / autocomplete menu, forcing users to create wrapper commands as a workaround
Installed plugin skills are not automatically loaded into Claude's context, meaning Claude doesn't auto-activate them or recognize they exist without exploration
Command-wrapping-skill pattern causes recursion: Creating a command that references a skill results in the skill re-triggering itself. Setting disable-model-invocation: true to prevent recursion also blocks the command from invoking the skill, creating a catch-22.