Plan mode feels read-only. Claude Code now blocks its edit tools until you approve a plan, but two gaps remain: sessions with bypass permissions available skip that block, and shell commands still run while planning. The instructions themselves are a short injected prompt - “MUST NOT make any edits” (Armin Ronacher, 2025-12-17) - and a March 2026 write-up got Claude to modify a .zshrc in plan mode. Writing MUST NOT in all caps is not a security boundary.
Make it actually read-only
Wire a PreToolUse hook that denies Write and Edit when permission_mode is plan, with one exception for the plan file under ~/.claude/plans/:
{
"hooks": {
"PreToolUse": [
{
"matcher": "Write|Edit|NotebookEdit",
"hooks": [
{
"type": "command",
"command": "~/.claude/hooks/block-plan-writes.sh"
}
]
}
]
}
}
The script reads the hook payload on stdin and exits with code 2 (blocking the call - exit code 1 is a non-blocking error) when the condition matches. This costs nothing in capability: the agent never legitimately writes anywhere else while planning.
For untrusted repos or unattended runs, climb the sandbox ladder instead. Decision tree for when to use plan mode: Plan mode decision tree. Full article: Claude Code plan mode.