jev-harness (Rust)
Zero-overhead System One decision harness & token optimizer for AI coding agents and Tauri applications.
jev-harness wraps TypeSafe Jev System One micro-decisions with local fast heuristics (< 500µs) and remote sub-second inference (70-300ms). It prevents catastrophic token waste ($10-$50/M frontier reasoning calls) by detecting dependency errors, circular failure loops, and deterministic routing locally.
⚡ Key Capabilities
-
Test & Process Failure Triage (
triage_test_failure):- Detects missing modules (
TS2307,Cannot find module,ModuleNotFoundError,E0463), flaky network timeouts (ETIMEDOUT,ECONNRESET), and syntax errors in < 500µs locally. - Tells your agent or process runner to install dependencies or retry without invoking expensive frontier LLMs (
skip_llm: true).
- Detects missing modules (
-
Loop & Doom Prevention (
should_abort_trajectory):- Evaluates consecutive identical test failures and repetition loops to kill runaway agentic runs before burning budget.
-
Dynamic Model Routing (
route_model_tier):- Routes simple tasks, typos, and lint errors to fast deterministic models, and reserves expensive 2026 reasoning models (GPT-6 Astra, Claude Fable 5.1 / Mythos 5.1) only for complex architectural asks.
-
Step Completion Verification (
verify_step_completion):- Confirms criteria satisfaction before concluding multi-step workflows.
📦 Installation
Add to your Cargo.toml:
[]
= "0.1"
= { = "1", = ["full"] }
Or install the standalone CLI:
🚀 Usage
1. Test Failure Triage
use triage_test_failure;
async
2. Trajectory Loop Abort Guard
use should_abort_trajectory;
async
3. Model Tier Routing
use route_model_tier;
async
4. CLI Usage
# Run triage on a traceback
# Trajectory abort check
# Route model tier
# Check system status