Skip to content

Disabled plugin skills still appear in system prompt (example-skills) #34940

Description

@lazariwilly

Bug Description

Plugin skills from disabled plugins still appear in the system prompt's skill list, consuming context window tokens even when explicitly disabled in .claude/settings.json.

Steps to Reproduce

  1. In .claude/settings.json, set a plugin to false:
    "enabledPlugins": {
      "example-skills@anthropic-agent-skills": false
    }
  2. Start a new Claude Code session
  3. Observe the system-reminder skill list

Expected Behavior

Skills from example-skills@anthropic-agent-skills should NOT appear in the system prompt when the plugin is disabled.

Actual Behavior

All ~15 example-skills:* skill descriptions (theme-factory, doc-coauthoring, claude-api, xlsx, pdf, algorithmic-art, internal-comms, skill-creator, canvas-design, pptx, slack-gif-creator, webapp-testing, frontend-design, mcp-builder, brand-guidelines, docx, web-artifacts-builder) appear in the system prompt alongside their enabled counterpart document-skills:*.

This means duplicate skill descriptions (document-skills + example-skills) consume context even though only document-skills is enabled.

Impact

Each skill description is ~50-150 tokens. With ~15 duplicated skills, that's ~750-2,250 unnecessary tokens in every conversation's system prompt. For users with many plugins, this compounds and contributes to a high baseline context usage (35%+ of window consumed before any user interaction).

Environment

  • Claude Code CLI (latest)
  • macOS Darwin 25.3.0
  • Settings file: .claude/settings.json with enabledPlugins configuration

Metadata

Metadata

Assignees

No one assigned

    Labels

    duplicateThis issue or pull request already exists

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions