Skip to content

Roadmap

These are candidates, not commitments. Each one names the gap in the current skills that it would close. The order is not a priority order.

design.md already states correctness properties, and spec-generator can add optional property-based test tasks, but no skill turns those properties into tests and runs them. A skill between the QA plan and implementation would make the properties executable.

spec-maintenance compares one new work item with the older ones, so it sees only the impacts that one item caused. It cannot see an old spec that no single change invalidated but several changes did together. Detecting that needs a check across many work items, not one run.

The audit travels with the work item on merge

Section titled “The audit travels with the work item on merge”

Nothing stops a work item from reaching main without its implementation-audit.md. The candidate is a merge rule, not a skill change: a work item directory merges with its audit, or the branch does not merge.

spec-qa-review plans five layers: unit, integration, end-to-end, performance and security. No layer covers accessibility.

The eval pack for spec-design-review already shows that broken designs return BLOCKED. The open gap is the other direction: a design with no real defect must reach APPROVED. The aligned clean-pass case now does, but the bugfix and prose-drift cases still return APPROVED WITH ISSUES. See #37.

The site is configured so that a ja locale can be added without changing the existing URLs.