How SuperX Code Works

SuperX Code operates through a structured pipeline: input expands into durable typed prompts; repository awareness produces a scoped context registry; agent planning produces bounded steps with permission guards; real tool execution produces structured results; results enter structured settlement before presentation.

Durable Prompts

User inputs promote into durable typed prompts before scheduling advisory execution wakes. Queued inputs promote at safe boundaries. Any promotion resets the agent's provider-turn allowance.

Project Understanding

The agent builds a location-scoped registry model of the repository โ€” files, modules, dependencies, relationships โ€” using observed context rather than full index scans.

Agent Planning

Planning breaks complex tasks into executable steps with bounded retry, permission checks, and failure reporting. The runner does not infer ambiguous provider work as safe to retry after promotion.

Execution

Real tool execution uses the secure registry. Output enters structured settlement. Oversized provider-facing text is bounded; structured metadata and media are preserved.

Validation

Results are validated before presentation. Failures produce operational errors rather than lossy successes.

Getting Started ยท Commands