The Agent Skills Directory

Command Execution Capability: The skill is designed to execute validation commands and perform file modifications to achieve a user-defined goal. This automated execution is the core functionality of the tool and is governed by a 'stop condition' to ensure the agent knows exactly when its task is complete.
Durable State Management: Progress and objective details are persisted in a session-specific file (goal.md). This allows the agent to maintain context over multiple interactions, providing a reliable source of truth for the ongoing task.
Safety Guardrails: To prevent unintended or infinite operations, the skill implements a max_checkpoints budget. This requires the agent to stop and consult the user once a certain number of steps have been taken, even if the goal hasn't been reached.
Defined Scope and Constraints: The instructions mandate a 'Constraints' section for every goal, explicitly documenting what the agent must not change. This practice helps ensure the agent operates within expected boundaries during autonomous work phases.

follow-goal