Repository navigation
Release v7.144.0
Summary
The reverse-engineering skill now scores how faithfully a model reproduces the target's behavior, not just its names, and requires stronger evidence before a model counts as validated.
Changed
- The reverse-engineering skill scores fidelity on observable behavior (rules, flows, rejections and time rules from a given/when/then catalog), reports naming fidelity separately, and discounts names the running app exposes verbatim. A new
score-fidelity.mjsscript computes the scores from a filled catalog (part of #1056) - The reverse-engineering skill makes reading the managed assemblies a standard intake step for owned .NET targets, and requires boundary and server-side rejection probes for every command, with a coverage gate before synthesis (part of #1056)
- The reverse-engineering skill seeds backdated data to find time-rule boundaries, and puts today-relative rules into query performer bodies that Stage renders instead of comments (part of #1056)
- The reverse-engineering skill requires every discovered specification to run before a model counts as validated, and covers API discovery, run-smoke routes, computed display values and pipeline boards (part of #1056)