SIGTREX API
AI guardrails for safer coding and agent execution
Start new thread
SIGTREX

SIGTREX API - AI guardrails for safer coding and agent execution

by•
SIGTREX API is an AI execution control layer for developers using ChatGPT, Claude, Cursor, Windsurf, Copilot, and AI agents. SIGTREX detects drift, blocks risky multi-step scope creep, generates safer prompts, creates execution contracts, verifies output before changes are applied, and creates replay traces. Example: Request: “Fix auth, redesign dashboard, and update billing” SIGTREX: 🚫 BLOCKED — multi_step_drift AI should follow scope — not quietly expand it.

Add a comment

Replies

Best
SIGTREX
Maker
📌
Built this after repeatedly running into the same problem: You ask AI for one thing. It quietly changes five. Touches auth. Changes billing. Redesigns UI. Modifies unrelated files. Working systems break. So I built SIGTREX API as an execution control layer between prompts and execution. The core idea: Prompt / Agent Guard → Safer Prompt → Execution Contract → Output Guard → Replay Trace Example: Request: “Fix auth, redesign dashboard, and update billing” SIGTREX: 🚫 BLOCKED — multi_step_drift Would genuinely love feedback from people building with ChatGPT, Claude, Cursor, Windsurf, Copilot, or agents.
Saul Fleischman

@sigtrex This is a real problem that compounds as teams scale—AI agents doing more than intended creates cascading failures that are hard to debug. The execution contract layer is smart because it forces explicitness rather than just hoping the model behaves. Have you found that guardrails need tuning per model, or does the approach work consistently across Claude, GPT, etc.