autopilot-scout — find loops & goals to set

autopilot-scout — find loops & goals to set

In my recent overnight build system review, I used the autopilot-scout tool to identify automation opportunities within our repository. The tool's purpose is to distinguish between recurring tasks and destination-driven tasks, providing paste-ready commands to help automate the right processes. This is essential for efficient continuous integration and delivery without unnecessary manual intervention.

Understanding /loop and /goal

The distinction between /loop and /goal is pivotal. The autopilot-scout defines /loop for recurring tasks that need regular attention, like CI babysitting or morning triage. These tasks are set with commands like /loop 10m <prompt>, which automatically execute at specified intervals. In contrast, /goal is for tasks with a clear endpoint, such as "all tests pass" or "build exits code 0". These tasks require a verifiable condition to be met, ensuring that the task is only completed when it meets this objective standard.

Building the Autonomy

To automate effectively, I started with the scan_autonomy.py script. It scans the repository for verifiable surfaces such as test, build, typecheck, and lint commands. The script produces autonomy.json, which ranks loop_candidates and goal_candidates with pre-written command strings. This method ensures that only tasks with clear, verifiable surfaces are automated, thus maintaining reliability and trust.

Qualifying Candidates

Once I had the candidates, I cross-referenced them with references/02-candidate-rubric.md, ensuring the cadence or verifiable surface matched the task's nature. For instance, if a task like "make tests pass" was identified as a goal, I ran the test_cmd to establish a baseline. This step is crucial because setting a goal is meaningful only if there is a current failure to address.

Proposing the Right Automation

Finally, I filled the assets/autonomy-report.template.md with a ranked shortlist of loop and goal candidates. This report provides a clear, actionable roadmap for automation. By focusing on tasks that genuinely need oversight or have a verifiable finish line, I ensure that our automated processes are both efficient and relevant.

Overall, autopilot-scout has proven indispensable in distinguishing between tasks that benefit from automation and those that do not. By setting clear criteria for looping and goal-setting, the tool helps maintain focus on meaningful automation, avoiding token wastage and ensuring trust in the system's autonomy.