Every retry loop, every hallucinated import, every "let me try a different approach" — those are billable tokens on every LLM vendor's bill. ThumbGate is machine-speed pre-action defense: thumbs-down once, block that exact mistake on every future call, surface the next highest-ROI remediation, and show which agent surfaces are actually active before rollout. Across Claude Code, Cursor, Codex, Gemini, Amp, Cline, OpenCode — any MCP-compatible agent, forever, including fast-moving vibe coding workflows.
As desktop agents move into parallel sessions, terminals, and production workflows, ThumbGate checks the thing benchmarks miss: is this next action a known workflow, an open-ended agent, a costly fan-out, or a blind tool call with no way to verify it worked?
Personal local dashboard · DPO export from real corrections · founder support on risky flows.
No credit card for 7-day trial · cancel anytime · your rules and captures stay local. Prefer free? Install CLI →
No, you do not have to chat inside the GPT forever. The GPT is advice and checkpointing; local hooks do the hard blocking for Claude Code, Cursor, Codex, Gemini, Amp, Cline, OpenCode, and MCP-compatible agents.
Your agent has no memory. Every session, the same wrong pattern runs. ThumbGate turns a single correction into a permanent block — before the next tool call fires. See all plans →
Prove one blocked repeat before asking anyone to buy. The fastest path to value: one person, one repeated mistake, one check that blocks it permanently.
Run npx thumbgate init in your repo. Or install the Claude Extension, Codex plugin, Cursor plugin, or open the GPT. Native ChatGPT rating buttons are not the ThumbGate capture path.
Give thumbs up when the agent follows your standards, or thumbs down when it misses. ThumbGate captures the context and distills a lesson from up to 8 prior entries.
Next time the agent tries the same mistake, the PreToolUse hook fires and physically blocks it. Upgrade after one real blocked repeat when you need the dashboard and exports.
This is the entire product in 2 lines. Install, give feedback, checks auto-generate.
Claude renders the live ThumbGate footer now. npx thumbgate init --agent codex installs the same Codex hook bundle and writes the ThumbGate statusLine target so you can test it on your local Codex build immediately.
ThumbGate ships a published Claude Desktop extension bundle (.mcpb) you can install today. Claude Code users can also add the repo marketplace plugin immediately. No waiting for directory approval.
Run npx thumbgate init --agent claude-code or add via claude mcp add thumbgate -- npx --yes --package thumbgate thumbgate serve
Download the .mcpb bundle for Claude Desktop, or use the repo marketplace: /plugin marketplace add IgorGanapolsky/ThumbGate
Type thumbs down when Claude makes a mistake. ThumbGate distills a lesson from up to 8 prior entries and blocks the pattern permanently via PreToolUse hooks.
Claude Code Skill: Type /thumbgate in any Claude Code session. Auto-triggers on “check”, “feedback”, “block mistake”. Free skill on top of the same local gateway.
ThumbGate should meet users where they already ask AI for help. The live GPT is the fastest way to preflight a risky action, capture a typed thumbs-up/down lesson, and prove the enforcement loop before installing anything. As ChatGPT ads roll out, this matters more: ChatGPT can stay the discovery and checkpointing layer, while ThumbGate remains the hard execution boundary after npx thumbgate init.
Paste a proposed command, file edit, merge, deploy, refund, invoice, or API call and ask whether to allow, block, or checkpoint it.
Reply in chat with thumbs up: or thumbs down: plus one concrete sentence. Do not rely on ChatGPT's native rating buttons for ThumbGate memory.
Run npx thumbgate init in the repo so Pre-Action Checks block repeated mistakes before the coding agent executes them.
Find it fast: if the direct link does not open, go to Explore GPTs, search ThumbGate, and choose the GPT by Igor Ganapolsky in Programming. Plain English rule: ChatGPT is the discovery and memory surface for advice, checkpointing, and typed feedback capture. One typed signal becomes one remembered rule. The hard Reliability Gateway still runs in the local agent or CI lane.
Install the published Claude Desktop plugin .mcpb bundle today. Claude Code users can add the repo marketplace immediately with /plugin marketplace add. No waiting for directory approval. 60-second setup guide →
Type /thumbgate in any Claude Code session. Auto-triggers on "check", "feedback", "block mistake". Free skill on top of the same local gateway teams later harden into a shared workflow.
Claude Code, Codex, Gemini CLI, Amp, and OpenCode all use the same gateway and memory model. Any MCP-compatible agent gets pre-action checks, feedback memory, and enforcement out of the box.
Cloud Next 2026 shipped BigQuery, Spanner, AlloyDB, and Cloud SQL as MCP tool calls into Claude Code, Codex, and Gemini CLI. ThumbGate checks the destructive ones — DROP on prod datasets, unscoped DELETEs, IAM escalation — before they fire.
Codex gets a standalone ThumbGate plugin bundle, a repo-local plugin profile, and the same auto-updating MCP launcher. The runtime resolves thumbgate@latest when Codex starts, so npm fixes reach active installs. The install page includes the zip, MCP config, and verification path in one place.
Drop the ThumbGate MCP config into .cursor/mcp.json and Cursor gets the same pre-action checks as Claude Code and Codex. Ships with bundled rules, commands, hooks, and agents.
VS Code works when you run an MCP-compatible agent inside it (Continue, Cline, etc.). Any editor that speaks MCP stdio gets the same gateway.
ThumbGate is listed on mcp.so so MCP-compatible clients can verify the package, copy the npx config, and confirm they are installing the real Pre-Action Checks server.
Open the ThumbGate GPT to preflight risky commands, deploys, refunds, PR actions, and setup steps, capture thumbs-up/down lessons, and save typed signals. Real blocking for coding agents still runs locally after npx thumbgate init.
Big orchestration suites unify data, routes, and decisions. ThumbGate sits closer to the moment of execution: the point where an agent is about to run a shell command, ship a PR, approve a release, or repeat a mistake you already corrected. That is where workflow trust is won or lost.
Good at customer journeys, routing, and cross-system context. Weak when you need a coding agent or automation to stop before a destructive or low-trust action actually runs.
Turns operator feedback into Pre-Action Checks. It does not just remember the mistake. It blocks the repeat at the tool-call boundary across Claude Code, Cursor, Codex, Gemini, Amp, Cline, and OpenCode.
Use orchestration to decide what should happen next. Use ThumbGate to decide what is allowed to execute. That is the control layer enterprises actually need once agents touch repos, terminals, CI, or production workflows.
Why IBM and Adobe-style orchestration still needs a stop layer once AI agents can touch code, approvals, or production workflows.
Standardize agent behavior across Claude Code, Codex, Cursor, Gemini, and CI without turning your rollout into an internal services project.
Keep approval boundaries, evidence, and audit context attached to the exact moment an agent wants to act in finance, healthcare, and other high-trust environments.
Why thumbs-up/down feedback and pre-action checks solve repeated mistakes faster than a manual spec-authoring loop.
Mem0 remembers context. ThumbGate enforces behavior. Why retrieval alone is not enough when your agent keeps repeating the same mistakes.
The core concept explained in plain language: how thumbs up, thumbs down, and runtime enforcement work together to prevent repeated failures.
Keep global prompts lean, load MCP schemas on demand, and turn harness lessons into Pre-Action Checks that block repeat failures.
Why AI tools recommend the brands they repeatedly see tied to a buyer problem, and how ThumbGate builds that association with proof-backed pages.
A long-tail answer-engine page for buyers asking how to gate Claude Code, Cursor, Codex, and parallel coding agents before risky execution.
As conversational AI becomes an ad surface, trust, measurement, and execution boundaries matter more. This page ties that shift directly to ThumbGate.
How LLMs store brand-to-problem associations, and why ThumbGate should own the “pre-action checks for AI coding agents” slot.
See how prompt injection, cross-app bridges, and silent connector installs turn browser-use into a governance problem, then audit what is already wired on disk.
Audit local browser bridges, missing host binaries, and pre-authorized extension paths before a desktop agent turns a one-off experiment into a durable integration.
Why structured feedback memory matters more when it becomes a live check that blocks your agent from repeating mistakes.
The shortest path from “I already corrected this once” to a pre-action check that blocks the repeat before the next tool call lands.
Why Cursor users need more than prompts when the same risky refactor, shell step, or git action keeps coming back.
Turn operator feedback into searchable lessons, linked prevention rules, and a runtime stop before Codex repeats the mistake.
Why Gemini CLI buyers start with memory and convert when they see how memory becomes real pre-action checks.
Why benchmark-search loops need checks for holdout tests, proof trails, reward hacking, and unsafe promotions.
Autoresearch loops run experiments, inspect metrics, and accept better variants. ThumbGate gives those loops a Reliability Gateway: Pre-Action Checks for skipped holdout tests, fake proof, reward hacking, unsafe edits, and promotion without verification evidence.
Require primary and holdout checks before an agent can call a variant better. Block cherry-picked runs and missing baselines.
Promotion needs commands, logs, changed files, and verification evidence so the win survives review instead of becoming a vague claim.
Start with templates for npm test, Playwright duration, bundle size, lint, and CI failures, then add Team checks for shared workflows.
Give 👍 or 👎 on your AI agent's actions. Feedback is stored in a SQLite+FTS5 lesson DB. In the current Claude auto-capture hook, a vague thumbs-down can distill from up to 8 prior recorded entries and the failed tool call before promotion, then stay linked to a 60-second feedback session. Example: you 👎 a risky migration → it auto-promotes to a "never run DROP on prod" check.
Repeated failures auto-promote into prevention rules. Thompson Sampling adapts which rules fire, and the reflector lane can propose a reusable rule from the same transcript so high-risk patterns get stricter enforcement while low-risk ones stay relaxed.
Rules become Pre-Action Checks that block your agent before it repeats the same mistake. Your agent can't force-push, skip tests, or repeat a refactor you already rejected. No more fix-loops.
Every block shows why: pattern match, evidence, confidence score.
Checks physically block tool calls. Not prompt tricks. Not "please don't."
Reasoning chains on every check decision. Thompson Sampling confidence tiers.
Repeated failures auto-promote to checks. Org dashboard shows all agents.
When Workflow Sentinel predicts a risky local action, ThumbGate can recommend Docker Sandboxes before the agent touches the host filesystem or broader credentials.
Changesets, SemVer, sync checks, and verification evidence make new package releases inspectable before a buyer trusts the next rollout.
npx thumbgate init --agent claude-code
Wires PreToolUse hooks automatically
npx thumbgate init --agent cursor
4 skills: feedback, rules, search, recall
npx thumbgate init --agent codex
6 skills including adversarial review
npx thumbgate init --agent gemini
Gemini CLI integration
npx thumbgate init --agent amp
Amp agent integration
npx thumbgate serve
MCP stdio server for any compatible client
Add to your claude_desktop_config.json:
{
"mcpServers": {
"thumbgate": {
"command": "npx",
"args": ["--yes", "--package", "thumbgate", "thumbgate", "serve"]
}
}
}
Official directory review is separate. Claude Code users can install immediately with /plugin marketplace add IgorGanapolsky/ThumbGate and /plugin install thumbgate@thumbgate-marketplace.
3 captures, 1 rule, 1 agent. Enough to prove the enforcement loop works. When you need more, you will know.
Unlimited captures, unlimited rules, full recall. $19/mo costs less than 20 minutes of re-fixing a mistake your agent already learned to avoid.
npm run export:hf)No credit card required. Cancel anytime. Your rules and captures stay local.
When one engineer teaches the agent not to delete staging data, that lesson applies to every agent on the team. Stop paying the same mistake tax across different developers.
team-import tagging built in. One team's hard-won lessons become every team's prevention rules$49/seat/mo with a 3-seat minimum. Start with a 30-minute intake around one real blocker.
This is the fastest path to first paid value for teams. Start with one workflow, one owner, and one blocker. The intake is designed to prove that ThumbGate reduces review churn, rollout risk, or repeated agent mistakes before a wider rollout.
npx thumbgate init.Join the mailing list for new check patterns, agent integration updates, and product news. If you later choose Pro, we keep checkout prefilled on this device.
Install free. No credit card. No signup. Hit your first check in 60 seconds.
See the enforcement before you buy anything
🔍 Live Dashboard Demo
Search lessons, inspect checks, mark a review checkpoint, and see only what changed since the last pass. No signup or install required.
Open Dashboard Demo →⛔ Check Reasoning Chains
Every block explains why: which pattern matched, what evidence triggered it, and whether the rule came from your own corrections.
📊 Org Dashboard (Team)
See which agents are creating review churn, which checks are saving time, and where rollout risk is still concentrated across the shared workflow.
🧱 Isolated Execution Lanes
High-risk local autonomy can route into Docker Sandboxes, while hosted team automations use a signed sandbox dispatch lane instead of running loose on a shared host.
🧪 Thompson Sampling
Confidence tiers (none/low/medium/high) tell you when to trust the model vs fall back to rules. No guessing.
🪞 History-Aware Lessons
When the current Claude auto-capture hook only gets a vague thumbs-down, ThumbGate can reuse up to 8 prior recorded entries plus the failed tool call, then keep a linked 60-second feedback session open for later corrections instead of creating a dead-end note.