Two parallel bash guard scripts use divergent naming conventions

[Widthdom]

Summary

There are two separate Bash guard scripts in this repo with divergent naming conventions: .codex/hooks/bash_guard.py (underscore) referenced from .codex/hooks.json:9, and .claude/hooks/bash-guard.py (hyphen) referenced from .claude/settings.json:152,183. CLAUDE.md and SELF_IMPROVEMENT.md document only the hyphenated variant. New maintainers must learn that there are two parallel guard scripts and that bash_guard.py (Codex side) is silently undocumented in the entry-point markdown. A drift between the two scripts (e.g., adding a banned word to one but not the other) would cause asymmetric enforcement across Codex vs Claude Code with no tooling to detect the divergence.

Where

• .codex/hooks.json:9 (references .codex/hooks/bash_guard.py)
• .claude/settings.json:152,183 (references .claude/hooks/bash-guard.py)
• .codex/hooks/bash_guard.py (underscore script)
• .claude/hooks/bash-guard.py (hyphen script)
• CLAUDE.md:19 (only mentions hyphen variant)
• SELF_IMPROVEMENT.md:44 (only mentions hyphen variant)

Suggested approach

1. Pick one canonical naming convention (hyphen or underscore) and rename one script to match — or document why two parallel scripts exist
2. If both scripts must remain as separate files, add a comment header in each pointing to the other and explaining the difference
3. Update CLAUDE.md and SELF_IMPROVEMENT.md to mention both scripts and clarify which fires under which agent
4. Add a small test or CI step that diffs the two scripts' banned-word lists and fails if they drift apart unintentionally
5. Document in DEVELOPER_GUIDE.md the policy: when modifying one guard, the maintainer must consider the other
6. Consider extracting shared rules (banned words, patterns) into a single source of truth that both scripts import

[Widthdom]

Work started on branch codex/fix-issue1998-1997-1995-1979-1906.