feat: add crit plan-hook for Claude Code PermissionRequest integration

[tomasz-tomczyk]

What changed

Adds a crit plan-hook subcommand designed to be used as a Claude Code PermissionRequest hook for ExitPlanMode. When Claude tries to exit plan mode, the hook intercepts it, opens a crit review session for the plan content, and blocks the exit until the user approves or adds comments.

• New crit plan-hook subcommand reads the PermissionRequest event JSON from stdin
• Extracts the plan content from tool_input.plan, saves it via the existing plan versioning system
• Starts a crit daemon (reusing the crit plan flow) and opens the browser for review
• Blocks until the user finishes reviewing, then outputs hookSpecificOutput JSON (allow/deny) to stdout
• All error paths fail-open (allow through) so hook failures never block the agent
• Updates the Claude Code plugin hooks/hooks.json to use crit plan-hook instead of the previous shell script (plan-review.sh), removing the $CLAUDE_PLUGIN_ROOT path dependency

Why

The previous shell script approach (plan-review.sh) relied on $CLAUDE_PLUGIN_ROOT being set by the plugin system, which was unreliable. Using the binary directly is simpler, has no path dependency, and gets native Go JSON parsing instead of jq.

Testing

# Build
go build -o crit .

# Simulate a hook event
echo '{"tool_input":{"plan":"# My Plan\n\nDo the thing."}}' | ./crit plan-hook

Entering plan mode in Claude Code and exiting should now trigger a crit review session.