[Claude Code Intro E30] PreToolUse Hook & Security: 4 Patterns Blocking Dangerous Commands
Many of the translated audio tracks on this channel are produced with HeyGen.
If you are interested in AI video translation, dubbing, and multilingual content creation, try HeyGen here:
https://bit.ly/4o4bn9P
Signing up through this link helps the channel - thank you!
While lesson 29 showed the PostToolUse that triggers after a tool finishes, lesson 30 is about PreToolUse, which triggers right before execution to prevent the call itself. It stops rm -rf before it goes to the shell and refuses .env edits – a security gateway guaranteed by the system, not the model's judgment.
📌 Topics Covered
Trigger condition: Right before tool execution — can block, modify, or allow the call. The exact opposite of PostToolUse.
Input JSON: Extracting tool_name, tool_input, tool_use_id with jq. No tool_output (since it hasn't run yet).
Two blocking methods: exit 2 + stderr vs exit 0 + permissionDecision JSON. The pitfall: exit 2 blocks, but exit 1 does not.
4 permissionDecision values and permissions system: allow (skips prompt) / deny (cancels) / ask (user confirmation) / defer (SDK only). The Hook only tightens constraints — deny beats bypassPermissions, but allow cannot override a deny rule.
4 Practical Patterns: Blocking dangerous Bash commands (rm -rf, DROP TABLE) / Protecting sensitive files (.env, .git/) / Approval escalation with ask / Input sanitization with updatedInput.
Common pitfalls: exit 2 blocking inversion, JSON is only for exit 0, vulnerability of static Bash argument patterns, shell profiles contaminating JSON output.
📌 Environment Info
Latest Claude Code version (Hooks system, PreToolUse event support)
.claude/settings.json (team sharing), .claude/hooks/*.sh (script files), CLAUDE_PROJECT_DIR environment variable
The if field (argument pattern matching) requires Claude Code v2.1.85+
External tool: jq
📌 Chapters
0:00 What is PreToolUse — the pre-execution checkpoint
0:28 Trigger conditions and comparing with PostToolUse
1:33 Input JSON and extracting fields with jq
2:50 Two blocking methods: exit 2 and exit 0 + JSON
4:19 4 permissionDecision decisions
5:08 Hook and permission system: Hook only tightens
7:26 Narrowing the matcher: matcher and if
8:26 Pattern 1: Blocking dangerous Bash commands
10:02 Pattern 2: Protecting sensitive files
11:30 Pattern 3: Approval escalation with ask
12:27 Pattern 4: Input sanitization with updatedInput
13:42 Common pitfalls and summary
📌 Useful Links
Blog: https://neosarchizo.github.io/posts/c...
Hooks official docs: https://code.claude.com/docs/en/hooks
Hooks guide: https://code.claude.com/docs/en/hooks...
Permissions official docs: https://code.claude.com/docs/en/permi...
Permission modes official docs: https://code.claude.com/docs/en/permi...
Settings official docs: https://code.claude.com/docs/en/settings
Bash command validation example: https://github.com/anthropics/claude-...
jq manual: https://jqlang.org/manual/
#ClaudeCode #PreToolUse #Security