learning

Learning

The adaptive memory that makes NXTG-Forge smarter every session -- capturing your preferences, patterns, and corrections so the system stops guessing and starts knowing.

Level	L1 Vibe Coder
Category	Governance & Analysis
Model	Haiku

---

What It Does

The Learning agent is NXTG-Forge's institutional memory. It observes how you work -- which agents you invoke, what corrections you make, what coding patterns you prefer -- and encodes those observations into persistent preferences that improve every future session.

Without this agent, every session starts from zero. You would repeat the same corrections ("I use Vitest, not Jest"), re-explain the same preferences ("conventional commits, always"), and watch the system make the same wrong guesses. The Learning agent eliminates that repetition by capturing signals from your behavior and making them available to all other agents.

It draws from three signal sources: session history (which agents run, in what order, how often), user corrections (when you override a recommendation or modify generated code significantly), and outcome tracking (did the generated code pass tests, was the commit accepted). The strongest signal is always an explicit correction -- when you say "I prefer X over Y," that preference is treated as authoritative.

When to Use It

• Repeated corrections: When you find yourself telling the system the same thing across multiple sessions -- "I use pnpm, not npm" or "I always want TypeScript strict mode."
• Recommendation quality is poor: When agent suggestions consistently miss the mark and you want the system to learn from your actual workflow patterns.
• Preference capture: When you want to explicitly set preferences for testing frameworks, commit styles, file organization, or code formatting that should persist.
• Pattern audit: When you want to understand what the system has learned about your workflow and verify it matches your actual preferences.