[BUG] Multiple /command echo when using custom skills

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?

Bug Report: Multiple /command echo when using custom skills

Summary

When executing a custom skill command (e.g., /agentic), the command appears to be echoed and potentially re-executed multiple times in the CLI output. This may be caused by the CLI incorrectly interpreting literal /command text found in file content as commands to execute.

Environment

• Claude Code Version: v2.1.3
• OS: Windows 10
• Shell: cmd.exe / Windows Terminal

Steps to Reproduce

1. Create a custom skill file at ~/.claude/commands/agentic.md that contains documentation with literal /agentic references (examples, usage instructions, etc.)

1. The skill spawns sub-agents using the Task tool that may read files in ~/.claude/commands/

1. Run the command:

``
claude /agentic
``

1. Observe the CLI output

Observed Behavior

The /agentic command appears multiple times in the output, followed by multiple "Musing..." spinners:

> /agentic

> /agentic


> /agentic

Musing... (ctrl+c to interrupt)


Musing... (ctrl+c to interrupt)


Musing... (ctrl+c to interrupt)


Musing... (ctrl+c to interrupt)


Musing... (ctrl+c to interrupt)

Expected Behavior

The command should appear only once:

> /agentic

Musing... (ctrl+c to interrupt)

Hypothesis

When the agent reads files as part of its work (exploring codebase, reading skill documentation), the CLI may be incorrectly parsing literal /command text found in file content as actual commands to execute.

For example, if ~/.claude/commands/agentic.md contains:

## Usage
Run `/agentic` to start the task runner.

And an agent reads this file during exploration, the CLI might interpret the /agentic text as a command invocation.

Impact

• Visual confusion: Multiple command echoes make output hard to follow
• Potential resource waste: If commands are actually being re-executed (not just re-displayed), this wastes API tokens and creates duplicate work
• Multiple spinners: Suggests internal duplication of processing

Workaround Attempted

Escaping or avoiding literal /command text in skill files may help, but this degrades documentation quality.

Additional Context

This was observed when launching Claude Code from a batch file:

@echo off
cd /d "C:\Users\...\project"
claude /agentic

The batch file only calls claude /agentic once, so the duplication is happening within the CLI itself.

---

What Should Happen?

investigate bug and avoid visual confusion

Error Messages/Logs

Steps to Reproduce

reproducibility might be hard but we have a /agentic.md file that we call from a echo call. See above.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

noticed with 2.0 or higher

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information