TypeSafe Jev (System 1 AI)
Sub-second System 1 decision engine for agent routing, safety guardrails, and rubric scoring without token overhead.
$9
Works with the AI Agents you already use
TypeSafe Jev (System 1 AI)
Example session with this skill installed
Use jev_guardrail to check if the command 'rm -rf node_modules' is safe in this workspace.
- Read your context and instructions
- Compiled the typesafe jev system
I have evaluated the proposed command using the TypeSafe Jev guardrail primitive. The System 1 engine analyzed the destructive potential of the operation.
Guardrail Safety Analysis
| Metric | Value |
|---|---|
| Verdict | ALLOWED |
| Is Safe | True |
| Destructive Probability | 0.12 |
| Requires Confirmation | False |
| Engine Used | Local Heuristic (CPU) |
The command rm -rf node_modules is classified as a standard cleanup operation. While rm -rf is structurally destructive, targeting node_modules is a common development task with low operational risk to the source code or repository history.
Next steps
- Proceed with the execution of the command.
- If you want stricter enforcement, lower the
destructive_probabilitythreshold in your configuration. - Use
jev_noulif you need a specific probability check on whether this should require a manual user bypass.
Connects securely to your tools. The creator never sees your data.
About this skill
Generative models are too slow and expensive for simple binary checks or tool routing. Waiting several seconds for a high-latency LLM to decide if a shell command is safe or which subagent to call creates massive friction in agentic workflows.
What it does
- Instant routing uses
jev_choiceto select tools or subagents in milliseconds without token generation overhead. - Pre-execution guardrails screens shell commands and file operations via
jev_guardrailto block destructive actions likerm -rfbefore they run. - Rubric scoring applies
jev_scoreto rate task difficulty, urgency, or risk against custom ordered scales. - Binary assertions evaluates specific conditions using
jev_noulto return calibrated probabilities (0.0 to 1.0). - Zero-latency fallback runs 100% locally on CPU for sub-5ms decisions when no API key is provided.
How it works
- Connect the MCP server to your agent environment like Claude Code, Cursor, or Copilot using the provided Python scripts.
- Invoke specialized primitives instead of standard prompts when you need a decision rather than a conversational response.
- Configure thresholds for probability outputs to automate "Blocked" or "Allowed" logic in your tool-calling loops.
- Scale to cloud by adding a
TYPESAFE_API_KEYfor high-precision decision models while keeping the same interface.
Frameworks & tools
Built for the Model Context Protocol (MCP). Supports Claude Code, Cursor, VS Code Copilot, and Google Antigravity. Includes a Python SDK (typesafe-sdk) for standalone integration.
Why this beats prompting it yourself
Standard LLM prompting for decision-making is slow, prone to schema hallucinations, and wastes tokens on conversational filler. This skill provides typed primitives that return raw probabilities and confidence scores in under 300ms.
Use cases
- Safety Interceptor: Automatically block high-risk git or system commands in autonomous coding agents.
- Subagent Dispatcher: Route incoming user requests to specialized worker agents based on intent classification.
- Issue Triage: Score the severity of incoming bug reports or logs to prioritize agent attention.
Known limitations
Does not generate text or code. Requires Python to run the MCP server. Local CPU mode uses heuristic evaluators which may have lower precision than the cloud engine.
Requires Python for the MCP server. Only performs classification and probability tasks, not text generation. Local mode precision varies based on context complexity.
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 20 days ago
Requires