Opt-in Add-on

agy-autotrain

Continuous empirical knowledge synthesis and runtime assumption verification for Antigravity peers.

Problem Solved

AI coding agent CLI harnesses change rapidly. New upstream updates to Google Antigravity frequently modify terminal formatting, prompt framing, headless print behavior, and shell syntax. Static hard-coded prompts quickly degrade or break.

agy-autotrain observes live interactions, captures operational anomalies, runs automated verification probes, and compiles verified behavioral facts into a clean, project-agnostic driver cheatsheet.

Main Capabilities

Autonomous Observation Capture: Tracks runtime friction (e.g. rate limits, formatting quirks, shell variances) during everyday work.
Empirical Probing: Executes targeted verification probes to confirm whether observed behavior is a consistent runtime contract.
Cheatsheet Compilation: Distills validated findings into driver-cheatsheet.core.md, optimizing agent context tokens.
Seed Synchronization: Propagates proven guidance to all Clavity driver instances without requiring code changes.

Installation & Setup

Install via the Claude Code marketplace:

claude plugin install agy-autotrain@clavity

Requirements & Prerequisites

Requirement	Specification
Host Plugin	Clavity Bridge (`clavity-dotnet` or `clavity-classic`)
Runtime Tools	PowerShell 7+ (for local probe verification workflows)
Supported Models	Gemini 3.1 Pro, Opus 5, and compatible peers

Known Limitations

Draining and curating captured knowledge requires human-in-the-loop maintainer approval.
Probes require an active Antigravity CLI session to measure live latency and output.

GitHub Destination

Read full documentation and examine the knowledge store in the repository: agy-autotrain on GitHub.