Document workaround for hidden explicit commands

[Kevinjohn]

Users can be told an explicitly entered /codex:* command is unavailable because Claude Code hides manual-only commands from the agent's skills list.

Closes #238.

Observable sequence

• Enter a command such as /codex:status or /codex:review.
• Expected: Claude attempts the explicit command.
• Actual: Claude may refuse because commands with disable-model-invocation: true are absent from its available-skills list.

Documentation change

• Add a troubleshooting entry linked to the upstream Claude Code issue.
• List every currently affected plugin command.
• Provide a ready-to-paste global CLAUDE.md instruction that distinguishes explicit user commands from automatic model invocation.

This is a documentation-only change; command configuration and runtime behavior are unchanged.

Validation

• node --test tests/commands.test.mjs: 8 passed
• Full test suite: passed
• git diff --check: clean

Actual screenshots

README workaround — actual terminal diff

Focused documentation validation — actual terminal output