ThumbGateThumbGate Verification evidence
ThumbGate vs claude-code-hooks

Copy-paste hooks on one laptop, or the same enforcement model across every agent and machine.

claude-code-hooks (by karanb192) is a hand-curated collection of bash hooks you drop into ~/.claude/hooks. It's a great free starting point. ThumbGate ships the same PreToolUse enforcement idea, but as a runtime that learns from your feedback, syncs lessons across machines, and works across Claude Code, Cursor, Codex, Gemini, Amp, Cline, OpenCode, and Claude Desktop without you maintaining shell scripts.

Both free at base tier Both PreToolUse-based Both local-first by default Different scopes

Side-by-side feature comparison

Capability claude-code-hooks ThumbGate
Distribution GitHub repo of shell scripts; copy-paste into ~/.claude/hooks npm package: npx thumbgate init wires hooks across every supported agent
Agents supported Claude Code only (uses Claude Code's native hooks API) Claude Code, Cursor, OpenAI Codex CLI, Google Gemini CLI, Sourcegraph Amp, Cline, OpenCode, Claude Desktop (via MCP)
Rule format Hand-written bash scripts (deny-lists, regex matchers) Auto-promoted from operator feedback through Thompson Sampling; precision/recall gated before any rule activates
Cross-machine sync Manual: copy the repo to each machine you use Pro tier: hosted lesson sync, so the same rule fires on laptop, CI, and teammate machines
Adapter maintenance You maintain it as Claude Code's hook API changes Maintained adapter matrix — when an agent runtime ships a breaking change, we update the adapter
Dashboard None (file-based logs) Hit rates, gate inventory, agent activity, DPO export, hosted at /dashboard
Telemetry / training data export None DPO preference pairs, HuggingFace dataset export, Databricks-friendly JSONL
License MIT MIT (npm package); Pro/Team are hosted services
Cost $0 forever $0 local CLI; $19/mo Pro for sync + dashboard; $49/seat/mo Team (min 3) for workflow hardening sprint

Pick claude-code-hooks if

  • You use Claude Code on one laptop and you're happy editing bash by hand.
  • You want every line of enforcement visible in a shell script you can grep through in 30 seconds.
  • You don't need the same rule to fire on a teammate's machine or on CI.
  • You haven't yet hit the "I've copied these hooks to my third laptop, this is getting silly" moment.

karanb192's repo is a legitimately great resource and the answers it provides for the basic blocklist questions are the right ones. We link to it from our docs. There's nothing about it that we're trying to disparage — it's just scoped to a smaller problem than what ThumbGate solves.

Pick ThumbGate if

  • You use more than one AI coding agent and you want the same rule to fire across all of them.
  • You don't want to babysit shell scripts when Claude Code, Cursor, or Codex ship a breaking change to their hook API.
  • You want lessons to auto-promote from operator feedback instead of writing every rule by hand.
  • You want a dashboard, exports, and the ability to share enforced rules across a team.
  • You're a law firm, finance team, or other regulated org that needs an audit trail of every blocked action.

Can I use both?

Yes. claude-code-hooks runs at Claude Code's native hooks layer. ThumbGate runs at the same PreToolUse boundary but is agent-agnostic and persists lessons across reinstalls and machines. They don't conflict. The most common dual-use pattern: borrow well-written hook ideas from claude-code-hooks (their block-dangerous-commands and protect-secrets sets are good seed material), let ThumbGate generalize those patterns into prevention rules that fire across every agent runtime you use.

FAQ

Is claude-code-hooks a direct ThumbGate competitor?

Same category, different scope. claude-code-hooks is a copy-paste bash hook collection for one laptop running Claude Code. ThumbGate is the same PreToolUse enforcement model packaged as a runtime engine with hosted sync, a multi-agent adapter matrix, and a dashboard. If you only need a few static rules on one machine, claude-code-hooks is fine. If you need the same rule to fire across machines, agents, or teammates, that's where ThumbGate fits.

Is ThumbGate's free CLI as capable as claude-code-hooks?

For the basic blocklist use case, comparable. ThumbGate's free tier ships 5 built-in checks plus the feedback loop and PreToolUse hook across 8 agent runtimes. claude-code-hooks ships more pre-written bash scripts you can install on day one. If you want to start with curated hand-written rules, install both — start with claude-code-hooks for the seed library, run ThumbGate alongside it to learn from your thumbs-down reactions.

What does ThumbGate Pro add over both the free tier and claude-code-hooks?

Hosted lesson sync, the maintained adapter matrix, the dashboard, DPO and HuggingFace export, and 24x7 ops on the rule engine. Pro is $19/mo. Team at $49/seat/mo (min 3) adds shared lesson DB and the Workflow Hardening Sprint engagement.

Where do I start?

If you want the smallest possible install: npx thumbgate init. If you want a curated set of hand-written hook examples to read: karanb192/claude-code-hooks. The two are complementary.