Skip to content

Release v3.54.0

Choose a tag to compare

@github-actions github-actions released this 12 Oct 21:15
00b1be1

The CLI catches up with the current Stage and Screenplay releases: cratis render admits models up to ESM v13, the embedded Screenplay MCP server carries model notes, and Copilot loads the Cratis AI instructions natively.

Added

  • cratis render admits ESM v11 (lossless Double-mode number literals), v12 (explicit enum member numbers) and v13 (concept compliance markers such as pii and secret). ESM v14 query filters stay refused with CLI-RENDER-004 until the bundled renderer admits them. (#332)
  • cratis screenplay mcp serves the model-notes tools from Screenplay 4.144: list-model-notes, propose-model-notes, read-proposal with view: "notes", and run-specifications with a proposalId to run a retained proposal's candidate model before apply. The reference explains that .screenplay/notes.json is owned by the server, committed with the model, and never blocks opening a model, and shows where to gate on open questions with screenplay check --open-questions report|fail. (#340)

Changed

  • Bundles Stage 4.79.1 and Screenplay 4.145.1, and runs cratis/stage:4.79.1 by default. Server-generated UUID values, scalar and named command responses, and event-triggered reactions now render instead of being refused.
  • cratis ai install and cratis ai update link .github/instructions to the corpus's generated Copilot adapters (.cratis/ai/harnesses/copilot/instructions), so Copilot applies the Cratis rules by path. An older corpus without those adapters keeps the link to .cratis/ai/rules, and cratis ai update relinks existing installs. (#341)

Fixed

  • cratis render --scope on a module or feature whose slices were all removed now checks only application-level declarations, so errors in unrelated slices no longer block removing the scope's previously rendered files. Application-level errors still refuse the render and leave the published output untouched. (#334)