@livx.cc/bare-v3 0.1.0-alpha.169 → 0.1.0-alpha.170
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/dist/manifest.json +1 -1
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -13,7 +13,7 @@ bun lab compare # comparison UI on localhost:4314
|
|
|
13
13
|
bun lab replay <run-id> # verifies saved app hashes, then serves its isolated preview
|
|
14
14
|
```
|
|
15
15
|
|
|
16
|
-
**Design system file.** `bun run design` generates `artifacts/bare-v3.design` (committed) from the real CSS with Para's img2design `html2design`: Cover, Foundations (tokens, light + dark), Components (one main component per `contracts/core.json` class, with the `data-b3-*` variants and states its CSS styles), Icons and Patterns (demo screens built from instances) — one board per page, every colour a token: light / dark are the `color` collection's modes, switched by Design's mode toggle (native-control pictures follow it as picture tokens). Components flow CSS → design only. Tokens flow both ways: edit them in Para Design, then `bun run design:tokens` writes the values back into `src/styles/foundation/tokens/*.css` and `bun run design` regenerates. `tests/design.test.ts` fails when the file is stale. Outside a sibling `bare-browser` checkout, set `IMG2DESIGN` to its `modules/design/skills/img2design/scripts/img2design.ts`.
|
|
16
|
+
**Design system file.** `bun run design` generates `artifacts/bare-v3.design` (committed) from the real CSS with Para's img2design `html2design`: Cover, Foundations (tokens, light + dark), Components (one main component per `contracts/core.json` class, with the `data-b3-*` variants and states its CSS styles), Icons and Patterns (demo screens built from instances) — one board per page, every colour a token: light / dark are the `color` collection's modes, switched by Design's mode toggle (native-control pictures follow it as picture tokens). Components flow CSS → design only. Tokens flow both ways: edit them in Para Design, then `bun run design:tokens` writes the values back into `src/styles/foundation/tokens/*.css` and `bun run design` regenerates. `tests/design.test.ts` fails when the file is stale. Each build also runs img2design `check`: Design's layout of every component against WebKit's own boxes — a new finding fails `bun run design` (known ones are listed with their reason in `CHECK_ALLOW`, `scripts/design.ts`). A whole new colour theme: Para Design's Theme (nothing selected → Theme: accent + neutral, previewed, Apply = one undo step; agent: `design.theme`), then `bun run design:tokens`. Outside a sibling `bare-browser` checkout, set `IMG2DESIGN` to its `modules/design/skills/img2design/scripts/img2design.ts`.
|
|
17
17
|
|
|
18
18
|
The comparison UI needs the fixture server for example links. Model outputs replay on separate local origins. Replay refuses missing or mismatched artifact hashes at startup and rechecks each served file against the recorded snapshot. Changed files return 409; unrecorded files are not served. Response headers `X-Bare-Artifact-Sha256` and `X-Bare-File-Sha256` identify the snapshot and exact served bytes without exposing run/strategy names. Run `bun scripts/capture-replay.ts <run-root> <run-id> <new-output-dir> [width height]` through the authorized browser driver to capture these network receipts, diagnostics and screenshot together. The output directory must be new and outside the original run. This initial-load capture leaves task/persistence unverified and does not certify later lazy-loaded resources; repair preparation must still validate the recorded binding against its intended parent. DOM content alone does not establish this binding. `bun lab verify-evidence evidence.json protocol.json` checks recorded measurements against the protocol viewport and verifies the referenced PNG file, size, and hash. For matched captures, declare `viewport.expectedPresentation` with `theme`, `rootFontSize` and `direction`; the browser probe records these values and missing or mismatched values fail validation. The theme measurement uses the explicit bare-v3 root attribute; other strategy adapters must map their own theme mechanism rather than guess. The probe also records `emptyActions`: buttons/links without visible text or a graphic candidate require review even when they have an accessible name. Fresh evidence must include this measurement; keep older captures unchanged. CSS-drawn affordances need adjudication, and a graphic candidate does not prove visible pixels or adequate contrast. Use `bun lab feedback evidence.json protocol.json` to emit a compact repair report with evidence/protocol fingerprints, screenshot verification and bounded findings. Like verification, it exits nonzero for unresolved findings or missing evidence; redirect stdout to retain the report. This is review-only and offline: it does not establish which app revision was captured, mutate artifacts, qualify the app, or start a paid edit run. For replay captures, append `--parent-run <run-directory>` to feedback or verify-evidence. This checks recorded document/resource receipts against the current parent files and manifest. It establishes consistency, not authenticity of an independently edited evidence file, and does not qualify the task. It does not judge visual quality or declare release qualification. Run artifacts and judgments live under ignored `artifacts/`; retain that directory when sharing evidence. Browser/model tooling is development-only and excluded from the packed consumer.
|
|
19
19
|
|
package/dist/manifest.json
CHANGED