ReAct Loop
A reasoning-and-acting agent loop where the model alternates "thought → action → observation" until the task is complete. VULK's generation agent uses a constrained ReAct loop with XML-tagged actions (vulkAction, vulkPlan, vulkCompletion).
ReAct Loop
ReAct (Reasoning + Acting) is the agent pattern introduced in the 2022 ReAct paper: the model produces a chain of thoughts and actions, each action calls a tool, the tool's output is fed back as an observation, and the loop continues until the model emits a stop signal. It is the standard control-flow for autonomous coding agents because it lets the model react to compile errors, file-system state, and search results rather than guessing in one shot.
VULK uses a constrained ReAct loop. The agent emits XML tags (vulkPlan, vulkAction type="file", vulkAction type="shell", vulkCompletion); each tag is parsed by the unified parser, the corresponding tool runs (write file, run shell, search knowledge base), and the result is appended to context. The loop terminates when vulkCompletion fires, when the 10-step quality gate passes, or when the budget guard (max tool calls, max tokens) trips.
Brand Engine
VULK's design-token and identity layer — 210+ pre-tuned brand profiles, 118 font pairings, and a color-forcing system that locks palette, typography, and motion before generation begins.
Quality Gate (10-step)
VULK's automated post-generation check — 10 sequential gates covering parse, install, typecheck, build, route resolution, asset integrity, accessibility, performance, and deploy readiness. A failure triggers either auto-fix or credit refund.