Claude on a Leash

Deterministic security guardrails that intercept Claude Code tool calls **before** they execute.

[](#install) [](#what-gets-installed) [](#mcp-server) [](LICENSE)

Claude Code can execute real actions in your repo. These hooks sit between Claude and your system, blocking dangerous commands, protecting secrets, and logging what happens. Six shell scripts. Zero trust.

---

Install

**Why not `curl | bash`?** One of these hooks *blocks* `curl | bash` patterns. We practice what we preach. Download the script, read it, then run it.

# Download
curl -fsSL https://raw.githubusercontent.com/adityaarakeri/claude-on-a-leash/main/install.sh \
  -o /tmp/claude-hooks-install.sh

# Inspect (please do this)
cat /tmp/claude-hooks-install.sh

# Install into current project
bash /tmp/claude-hooks-install.sh

# OR: install globally for all repos
bash /tmp/claude-hooks-install.sh --global

Other options

bash install.sh --dry-run               # preview what would be installed
bash install.sh --uninstall             # remove leash hooks (project-local)
bash install.sh --uninstall --global    # remove leash hooks (global install)
bash install.sh --no-color              # plain output (for CI)

---

What gets installed

Claude Code

Six hooks wired via `.claude/settings.json`:

Hook file	Event	Matcher	What it stops
bash-safety-guard.sh	PreToolUse	Bash	Destructive commands, RCE, exfiltration, reverse shells, privile