The governed pipeline

DevAware OS treats AI execution as a six-stage controlled pipeline. Each stage gates the next; nothing crosses from thinking to doing by accident.

The six stages

1. Intent. The developer states, in natural language, what this loop should accomplish. A P0 filter blocks low-signal intents before they consume a cycle.
2. Governance. Governance blocks decide which tools may run, what file scope is in bounds, and whether an action requires explicit approval.
3. Clean prompt. The prompt weaver assembles a scoped, secret-free prompt with explicit acceptance criteria. The context modal lets you inspect it before sending.
4. Controlled execution. Tool calls run through an allowlist; destructive commands are refused at the spawn boundary; flagged actions pause for approval.
5. Validation. Typecheck, lint, test, build, and smoke gates must pass before the loop declares work done.
6. Compact memory. Outcomes are distilled into a local store the next loop can retrieve from.

Why a pipeline

An AI coding agent without a control layer is convenient and unsafe. The pipeline makes the agent's permitted actions explicit, inspectable, and reversible — without slowing the human down on the thinking side.

See also: How it works, Governance blocks.