# Malformed flag `-dangerously-skip-permissions` (single dash) silently enables `--debug` via prefix match

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?

● # Malformed flag -dangerously-skip-permissions (single dash) silently enables --debug via prefix match

## Summary
Passing -dangerously-skip-permissions (single dash, typo) to the claude CLI
does not error out. Instead, the arg parser matches the leading -d against
the --debug shorthand, silently enables Debug mode, and discards the rest of
the token. --dangerously-skip-permissions itself is never applied.

This is dangerous for two reasons:

1. Security-relevant flag silently ignored. The user believes they are

running with permissions disabled; they are not. The opposite would be worse
(thinking permissions are enforced while they're actually off), but either
direction is a trust violation.

1. Debug mode activates without consent. Verbose logs (including MCP

traffic, tool I/O, hook execution, and request metadata) are written to
~/.claude/debug/<session>.txt without the user asking for them.

## Repro
```bash
claude -dangerously-skip-permissions

1. CC launches normally.
2. Statusline shows "Debug mode" indicator.
3. ~/.claude/debug/<session-uuid>.txt starts filling with [DEBUG] entries.
4. Permission prompts still appear — --dangerously-skip-permissions was not

honored.

Expected

Unknown / malformed flags should exit non-zero with a clear error, e.g.:
error: unknown option '-dangerously-skip-permissions'
did you mean '--dangerously-skip-permissions'?

Actual

• Exit code: 0 (CC runs).
• Debug mode: on.
• --dangerously-skip-permissions: not applied.
• No warning printed.

Root cause (suspected)

The CLI arg parser appears to use non-strict prefix matching on single-dash
short options: -d matches --debug, and anything after -d is swallowed
as the attached value or dropped. A strict parser (e.g. clap with
allow_hyphen_values = false or Node parseArgs({ strict: true })) would
reject the token.

Impact

• Low, since the typo fails safe (permissions stay enforced).
• Becomes high if an analogous typo ever maps to a flag that weakens

security instead of logging. The underlying permissiveness is the real
issue; this specific instance is a harmless symptom of it.

Suggested fix

Reject unknown options. If short-flag prefix matching is intentional, at
minimum emit a warning when an unrecognized long-looking token (-xxxxx+)
starts with - and is > 2 chars, and the matched short flag would consume
unexpected trailing characters.

Environment

• OS: Windows 11 Pro (26200)
• Shell: bash (Git for Windows)
• CC version: 2.1.112 (native installer, ~/.local/bin/claude.exe)

What Should Happen?

Expected

Unknown / malformed flags should exit non-zero with a clear error, e.g.:
error: unknown option '-dangerously-skip-permissions'
did you mean '--dangerously-skip-permissions'?

Error Messages/Logs

Actual

  - Exit code: 0 (CC runs).
  - Debug mode: on.
  - --dangerously-skip-permissions: not applied.
  - No warning printed.

Steps to Reproduce

Repro

```bash
claude -dangerously-skip-permissions

1. CC launches normally.
2. Statusline shows "Debug mode" indicator.
3. ~/.claude/debug/<session-uuid>.txt starts filling with [DEBUG] entries.
4. Permission prompts still appear — --dangerously-skip-permissions was not

honored.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.112 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_