Works with the AI Agents you already use
thumbgate pretool gate
Hard-block secret leaks and recurring agent mistakes using a pre-tool execution gate.
$9
thumbgate pretool gate
Example session with this skill installed
The agent is about to push a file containing an API key to a public gist. Use ThumbGate to check this action.
- Read your context and instructions
- Compiled the thumbgate pretool gate
I have intercepted the tool call to prevent a potential secret leak. The action has been blocked based on the ThumbGate pre-tool security policy.
Security Gate Interception Report
| Metric | Status | Detail |
|---|---|---|
| Action | DENIED | Attempted curl to gist.github.com |
| Violation Type | Secret Exfiltration | High-entropy string detected in payload |
| Rule Source | Local Gate Policy | secrets-pretool |
| Turn Status | Terminated | Execution halted to protect credentials |
Gate Feedback Log
{
"feedback": "up",
"context": "Agent attempted to upload file containing sensitive API keys to a public endpoint.",
"what-worked": "Hard block triggered on tool input scan",
"tags": ["secrets", "pretool"]
}
Next steps
- Run
npx thumbgate@1.37.2 gate-statsto see the total number of blocked exfiltration attempts. - Remove the sensitive keys from the source file or move them to a
.envfile excluded from the agent's scope. - Use
npx thumbgate@1.37.2 feedbackto document any specific context for this block.
Connects securely to your tools. The creator never sees your data.
About this skill
Standard LLM agents often leak environment secrets or repeat documented mistakes despite system prompts. This skill integrates ThumbGate into your agent's execution loop to provide a hard block on sensitive data exfiltration and recurring failures. It enforces a strict "deny-by-default" policy for secrets and ranked lessons, preventing the agent from proceeding with risky tool calls.
What it does
- Enforce secret-leak blocks by scanning tool commands, URLs, and file paths for sensitive data.
- Prevent recurring failures by checking tool calls against a local registry of ranked lessons.
- Scaffold gate infrastructure using
npx thumbgate initto set up local state and MCP configurations. - Track gate performance by recording "thumbs-down" feedback for misses and "thumbs-up" for successful blocks.
- Report gate statistics to provide a clear view of how many risky actions were prevented.
How it works
- Initialize the gate within the project directory to establish the local security state and MCP config.
- Execute pre-tool checks by piping tool calls through the gate-check command before execution.
- Capture feedback after an action to train the local model on what to block or what was missed.
- Monitor health by reviewing counters and stats to ensure the security policy is firing correctly.
Frameworks & tools
This skill utilizes the thumbgate npm package (v1.37.2) and integrates via npx commands. It is compatible with projects using Node.js and agents supporting Model Context Protocol (MCP) or CLI-based tool interception.
Why this beats prompting it yourself
Prompt-based instructions to "not leak secrets" are frequently ignored by models under pressure. This skill moves the logic into a hard execution gate that physically stops the tool call, ensuring that a warning results in a turn termination rather than a retry.
Use cases
- Prevent agents from accidentally posting
.envsecrets to external API endpoints. - Block the agent from repeating a specific coding pattern that previously failed a test.
- Integrate a local security firewall into Claude Code or Cursor workflows.
- Audit agent reliability by tracking how often the gate intercepts a failure.
Known limitations
Does not read home-directory credentials or environment variables. Does not replace branch protection or human review. Hosted Pro features like shared history are not included.
Requires Node.js for npx execution. Does not access home-directory credentials. Hosted Pro features (shared dashboard, proof) are not included in this local-first version.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Requires