codex-goal-loop

A Codex goal loop is a contract-driven workflow that automates persistent autonomous agents using plan-act-test-review iterations. It governs long-running tasks through a 4-part contract and lifecycle states, turning vague prompts into verifiable executions with clear stop conditions.

To start a persistent goal loop, configure Codex in your working directory and issue a /goal contract. You must define the objective, constraints, validation criteria, and stop condition to govern the autonomous agent's multi-step refactoring automation.

Yes, you can pause and resume autonomous agents during complex migrations. The skill tracks lifecycle states including pursuing, paused, achieved, unmet, and budget-limited, allowing you to apply start and stop controls to manage checkpoints.

The 4-part contract consists of an objective, constraints, validation, and a stop condition. These components work together to govern long-running agent workflows, ensuring persistent executions remain verifiable and stop-condition-guided across software environments.

Use a contract-driven goal loop for long-running migrations, feature benches, or codebase refactors requiring multi-step automation with checkpoints. Standard automation prompts lack the lifecycle states and 4-part contracts needed to manage verifiable, persistent agent executions.

Lifecycle states track progress as pursuing, paused, achieved, unmet, or budget-limited, but the approach requires defining strict validation and stop conditions. Without clear multi-step checkpoints, complex projects may hit budget limitations before the objective is achieved.