goal-finder — derive the goals a repo needs to finish

Practical Goal Derivation with goal-finder

Last night, our build system ran the goal-finder tool on the Fleet Control Plane repo. The output was clear and actionable, which is exactly what I hoped for from this skill. goal-finder takes a repository's scattered documentation and synthesizes it into a list of verifiable goals — a true "definition of done."

How the Modes Work

The tool operates in two modes: Mode A for a repo-wide backlog and Mode B for a contextual next-goal. In our use case, Mode A was triggered. This mode scans all documentation, including roadmaps and unchecked task lists, to produce a coherent backlog. The backlog isn't just a to-do list; it orders tasks by build order and prioritizes them based on impact and dependencies.

Why goal-finder Matters

The beauty of goal-finder is its focus on creating goals as verifiable contracts. Unlike vague objectives, each goal specifies a completion condition, such as a test passing or a command exiting with a status of 0. This is crucial because it turns abstract tasks into concrete, achievable milestones.

The Output

The output was a set of 15 goals, with each goal represented in a single paragraph. This format is intentional; it ensures that each goal is self-contained and repo-specific, making it easy to copy-paste and execute in a /goal run or a focused work session.

Practical Rules

One of the most useful aspects of goal-finder is how it enforces operating rules. For example, every task gets written to a shared checklist, following .claude/skills/goal-autopilot/references/SHEET-AUTHORING-CHECKLIST.md. This ensures consistency and prevents tasks from falling through the cracks.

Goals are grouped into phases that can run in parallel, provided there are no unmet prerequisites. This parallelization can drastically reduce the time it takes to complete a project, a feature especially valuable in complex repos like the Fleet Control Plane.

Compute Class

Each task is assigned a compute class between 1 and 10, indicating the level of computational effort required. For instance, a task that involved cross-system coordination was rated as a compute class 7, which denotes a high complexity but manageable within ordinary build/test efforts.

Conclusion

The real power of goal-finder lies in its ability to transform a chaotic set of documents into a clear, actionable roadmap. By breaking down tasks into verifiable contracts, it not only clarifies the path to completion but also makes it more achievable. This is not just about knowing what needs to be done; it's about knowing how to get it done efficiently.