Repository navigation
Release v7.49.0
Summary
The Screenplay and event-modeling skills improve the modeling method, with the most attention on turning existing code bases into evidence-backed Screenplay models.
Added
- Legacy extraction reads tests as evidence for rules, error messages, permissions and example data, and re-runs read only what changed since each feature was last analyzed, which also checks that an agreed freeze held. (part of #672)
- Legacy extraction works breadth first: a whole-system outline, then one feature per pass with a detail budget, steered by the domain expert; features not yet extracted are listed in the loss report. (part of #672)
- The extraction evidence keeps a decision trail, a code-term to business-term glossary and where each feature lives in the code, and covers generic engines, reference data, event order and outbound outcomes. Experts verify through plain-language retellings and a keep, change, new or remove view per slice. (part of #672)
- Discovery marks how certain each item taken from a document is, keeps numbers and dates exact, and records contradictions as questions. (part of #672)
- Command cards name the values each rule reads, field lineage is checked while editing, and a business-rule test decides which rejections and duplicates earn a specification. (part of #672)
- Model review offers an optional stressor analysis and a trace of one piece of information through the whole model. (part of #672)
- Render and gap-fill checks hand-written code that changed since it was last verified against the accepted model, and keeps a per-slice record of where each declaration lives in code. (part of #672)
- The Slice Implementer looks for an existing realization before writing one, mirrors the nearest delivered slice, and records non-behavioral assumptions instead of stopping on them; slice specifications explain how to turn a
.playspecification into C# specs. (part of #672) - The write guard hook stops agents from editing their own settings,
.mcp.json, hook configuration and, in repositories with an installed corpus, the managed.cratis/aifiles and installed adapters. (part of #672)