ThumbGate

thumbgate

ThumbGate provides pre-action gates for AI coding agents. It captures thumbs-up/down feedback on agent actions, auto-promotes repeated failures into prevention rules, and blocks known-bad tool calls via PreToolUse hooks. Trigger when the user wants to add safety guardrails to an AI agent workflow...

IgorGanapolsky/mcp-memory-gateway·0installs27stars

SKILL.md

Full skill instructions

ThumbGate

Pre-action gates that stop AI coding agents from repeating known mistakes.

Quick Start

npx thumbgate init

This installs the MCP server and wires it into your agent's tool configuration. No API keys required for the free tier.

Or install globally:

npm install -g thumbgate
thumbgate init

MCP Configuration

Add to your agent's MCP config (e.g., claude_desktop_config.json or .cursor/mcp.json):

{
  "mcpServers": {
    "thumbgate": {
      "command": "npx",
      "args": ["-y", "thumbgate"]
    }
  }
}

How It Works

Feedback Capture

When an agent action succeeds or fails, capture structured feedback:

Thumbs up: Records what worked, tags it, and stores it as a reusable pattern.
Thumbs down: Records the failure context, what went wrong, and what to change. Repeated failures auto-promote into prevention rules.

Prevention Rules

After a failure pattern repeats (default: 2 occurrences), ThumbGate auto-generates a prevention rule. These rules are injected into agent context before every tool call, blocking the known-bad pattern before it executes.

Pre-Action Gates

Gates intercept tool calls via the MCP PreToolUse hook. Each gate checks the proposed action against:

Prevention rules generated from past failures
Thompson Sampling confidence scores (adaptive sensitivity)
LanceDB vector similarity to known-bad patterns

If a match is found, the gate blocks execution and surfaces the prior failure context.

Context Packs

Bounded retrieval of relevant feedback history for the current task. The agent gets exactly the lessons that matter, not the entire history.

MCP Tools Provided

Tool	Description
`capture_feedback`	Record thumbs-up/down with structured context
`search_lessons`	Query the lesson DB by keyword or semantic similarity
`get_prevention_rules`	Retrieve active prevention rules for the current context
`session_primer`	Get session handoff context (last task, next step, blockers)
`export_dpo`	Export feedback pairs in DPO training format

Tiers

	Free	Pro	Team
Feedback capture	3/day	Unlimited	Unlimited
Lesson search	5/day	Unlimited	Unlimited
Active gates	5	Unlimited	Unlimited
Dashboard	-	Yes	Yes
DPO export	-	Yes	Yes
Seats	1	1	Per-seat
Price	$0	$19/mo or $149/yr	$49/seat/mo

Start a 7-day free trial of Pro: https://thumbgate-production.up.railway.app/pro?utm_source=skill&utm_medium=agent_context&utm_campaign=pro_page

Compatibility

ThumbGate works with any MCP-compatible agent:

Cursor
Codex
Gemini CLI
Amp
OpenCode
Any agent supporting the Model Context Protocol

Install skill

Copy and run in your terminal

npx -y skills add https://github.com/IgorGanapolsky/mcp-memory-gateway --skill thumbgate

View on GitHub

SKILL.md

adversarial-review
Run a skeptical Codex review from Claude Code that hunts for hidden regressions, unsafe assumptions, and release risk. Use before merges, deploys, migrations, or other risky changes.
0 installs
prevention-rules
Generate and review prevention rules auto-promoted from repeated failure patterns.
0 installs
setup
Validate that the Codex bridge plugin is ready inside Claude Code. Use when the user asks to install, configure, or sanity-check Codex review from Claude.
0 installs
solve-architecture-autonomy
Course-correct from over-planning to execution — when the agent has produced a plan, research, or options instead of doing work the user already approved, drop the analysis and execute the smallest end-to-end action now (branch, run, deploy, verify). Use when the user says "stop planning, just do...
0 installs