Enforce Hooks

Enforce Hooks

Generate PreToolUse hook scripts from CLAUDE.md directives.

When to use

• User says "enforce my CLAUDE.md rules" or "generate hooks from my CLAUDE.md"
• User complains about Claude ignoring CLAUDE.md instructions
• User wants code-level enforcement of their rules
• User mentions @enforced directives

How it works

1. Read the user's CLAUDE.md (or the file they specify)
2. Classify each directive as enforceable or not
3. For each enforceable directive, generate a standalone bash hook script
4. Show the user what you found and what hooks you would create
5. On confirmation, write the hooks and update .claude/settings.json

What is enforceable

A directive is enforceable if it can be checked at tool-call time. The hook receives the tool name and input JSON before execution and can block it.

Enforceable (yes):

• "Never modify .env files" -> file-guard hook (block Write/Edit to .env)
• "Don't force push" -> bash-guard hook (block push --force in Bash)
• "Always search locally before using web search" -> require-prior-tool hook
• "Don't commit to main" -> branch-guard hook (block git commit on main)
• "Never run rm -rf" -> bash-guard hook (block pattern in Bash commands)
• "Don't edit files in vendor/" -> file-guard hook (block by path pattern)
• "Always run tests before committing" -> require-prior-tool hook
• "Never use sudo" -> bash-guard hook
• "Don't read files in secrets/" -> file-guard hook (block Read too)
• "Use TypeScript, not JavaScript" -> file-guard hook (block Write to *.js)

Not enforceable (skip with explanation):

• "Write clean code" (subjective, no tool-call signal)
• "Use descriptive variable names" (code quality, not a tool constraint)
• "Follow REST conventions" (architectural, not checkable at tool-call time)
• "Be concise in responses" (output style, not tool usage)

Hook templates

Use these templates as the basis for generated hooks. Each hook is a standalone bash script that reads JSON from stdin and outputs a JSON decision.

Template: file-guard

Blocks Write, Edit, or Read operations on protected files.

#!/usr/bin/env bash
# enforce: {DIRECTIVE_TEXT}
# Generated from CLAUDE.md by enforce-hooks skill
INPUT=$(cat)
TOOL=$(echo "$INPUT" | jq -r '.tool_name')
case "$TOOL" in Write|Edit|MultiEdit|Read) ;; *) exit 0 ;; esac
FILE=$(echo "$INPUT" | jq -r '.tool_input.file_path // empty')
[ -z "$FILE" ] && exit 0
BLOCKED_PATTERNS=({PATTERNS})
for pat in "${BLOCKED_PATTERNS[@]}"; do
  [[ "$FILE" == *"$pat"* ]] && echo "{\"decision\": \"block\", \"reason\": \"Protected: $FILE matches $pat. CLAUDE.md: {SHORT_DIRECTIVE}\"}" && exit 0
done
exit 0

Replace {PATTERNS} with quoted glob patterns like ".env" "secrets/" "*.pem". Replace {DIRECTIVE_TEXT} with the original CLAUDE.md text. Replace {SHORT_DIRECTIVE} with a truncated version for the block message. Adjust the case statement: remove Read if the directive only blocks writes.

Template: bash-guard

Blocks dangerous patterns in Bash commands.

#!/usr/bin/env bash
# enforce: {DIRECTIVE_TEXT}
INPUT=$(cat)
[ "$(echo "$INPUT" | jq -r '.tool_name')" != "Bash" ] && exit 0
CMD=$(echo "$INPUT" | jq -r '.tool_input.command // empty')
BLOCKED=({PATTERNS})
for pat in "${BLOCKED[@]}"; do
  [[ "$CMD" == *"$pat"* ]] && echo "{\"decision\": \"block\", \"reason\": \"Blocked command pattern: $pat. CLAUDE.md: {SHORT_DIRECTIVE}\"}" && exit 0
done
exit 0

Replace {PATTERNS} with quoted command fragments like "push --force" "rm -rf /" "sudo ".

Template: branch-guard

Blocks git operations on protected branches.

#!/usr/bin/env bash
# enforce: {DIRECTIVE_TEXT}
INPUT=$(cat)
[ "$(echo "$INPUT" | jq -r '.tool_name')" != "Bash" ] && exit 0
CMD=$(echo "$INPUT" | jq -r '.tool_input.command // empty')
BRANCH=$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "unknown")
PROTECTED=({BRANCHES})
for b in "${PROTECTED[@]}"; do
  if [ "$BRANCH" = "$b" ]; then
    case "$CMD" in
      *"git commit"*|*"git merge"*|*"git push"*)
        echo "{\"decision\": \"block\", \"reason\": \"Branch $b is protected. CLAUDE.md: {SHORT_DIRECTIVE}\"}"
        exit 0 ;;
    esac
  fi
done
exit 0

Replace {BRANCHES} with quoted branch names like "main" "master" "production".

Template: require-prior-tool

Blocks a tool unless another tool was used first in the session.

#!/usr/bin/env bash
# enforce: {DIRECTIVE_TEXT}
INPUT=$(cat)
TOOL=$(echo "$INPUT" | jq -r '.tool_name')
# Only check for the target tools
case "$TOOL" in {TARGET_TOOLS}) ;; *) exit 0 ;; esac
# Check session log for required prior tool
TODAY=$(date -u +%Y-%m-%d)
LOG="$HOME/.claude/session-logs/$TODAY.jsonl"
if [ -f "$LOG" ]; then
  if grep -q '"tool":"{REQUIRED_TOOL}"' "$LOG" 2>/dev/null; then
    exit 0
  fi
fi
echo "{\"decision\": \"block\", \"reason\": \"Run {REQUIRED_TOOL} first. CLAUDE.md: {SHORT_DIRECTIVE}\"}"
exit 0

Replace {TARGET_TOOLS} with pipe-separated tool names like WebSearch|WebFetch. Replace {REQUIRED_TOOL} with the tool that must be used first like Grep.

Note: require-prior-tool needs session-log hook installed to work. If it's not installed, mention this to the user and offer to set it up.

Template: tool-blocker

Unconditionally blocks a specific tool.

#!/usr/bin/env bash
# enforce: {DIRECTIVE_TEXT}
INPUT=$(cat)
TOOL=$(echo "$INPUT" | jq -r '.tool_name')
case "$TOOL" in {BLOCKED_TOOLS}) ;; *) exit 0 ;; esac
echo "{\"decision\": \"block\", \"reason\": \"{BLOCK_REASON}. CLAUDE.md: {SHORT_DIRECTIVE}\"}"
exit 0

Installation procedure

After generating hooks, install them:

1. Create directory: mkdir -p .claude/hooks
2. Write each hook script to .claude/hooks/{hook-name}.sh
3. Make executable: chmod +x .claude/hooks/{hook-name}.sh
4. Update .claude/settings.json to register the hooks

The settings.json hooks section should look like:

{
  "hooks": {
    "PreToolUse": [
      {"type": "command", "command": ".claude/hooks/enforce-file-guard.sh", "timeout": 5000},
      {"type": "command", "command": ".claude/hooks/enforce-bash-guard.sh", "timeout": 5000}
    ]
  }
}

Each hook runs on every tool call and does its own filtering internally.

If .claude/settings.json already exists with hooks, MERGE the new hooks into the existing array. Do not overwrite existing hooks.

Interaction pattern

Present findings as a table:

Found N enforceable directives in your CLAUDE.md:

| # | Directive | Hook type | What it blocks |
|---|-----------|-----------|----------------|
| 1 | "Never modify .env" | file-guard | Write/Edit to .env* |
| 2 | "Don't force push" | bash-guard | push --force, push -f |
| 3 | ... | ... | ... |

Skipped M directives (not enforceable at tool-call level):
- "Write clean code" (subjective, no tool-call signal)
- ...

Generate these hooks? I'll create them in .claude/hooks/ and update settings.json.

Wait for user confirmation before writing any files.

Testing

After installation, suggest the user test each hook:

To verify your hooks work:
1. Try editing .env -> should be blocked with a clear message
2. Try running `git push --force` -> should be blocked
3. Normal operations should work unchanged

Source: Bande-a-Bonnot/Boucle-framework — distributed by TomeVault.