@ggui-ai/ui-gen 0.19.0 → 0.21.0
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/dist/adapters/base.d.ts +7 -0
- package/dist/adapters/base.js +1 -0
- package/dist/adapters/base.js.map +1 -1
- package/dist/adapters/generation-dispatch.d.ts +17 -3
- package/dist/adapters/generation-dispatch.js +992 -3468
- package/dist/adapters/generation-dispatch.js.map +1 -1
- package/dist/adapters/index.js +70 -13
- package/dist/adapters/index.js.map +1 -1
- package/dist/adapters/types.d.ts +1 -1
- package/dist/adapters/types.js.map +1 -1
- package/dist/advanced/index.d.ts +4 -4
- package/dist/advanced/index.js +1060 -3520
- package/dist/advanced/index.js.map +1 -1
- package/dist/blueprint-validator.js +13 -3
- package/dist/blueprint-validator.js.map +1 -1
- package/dist/boilerplate.js +2 -2
- package/dist/boilerplate.js.map +1 -1
- package/dist/check/index.d.ts +1 -1
- package/dist/check/index.js.map +1 -1
- package/dist/{create-ui-generator-CxPvudbs.d.ts → create-ui-generator-D7kOXNjJ.d.ts} +2 -2
- package/dist/{dispatch-N8wdNOX-.d.ts → dispatch-DGvhmARp.d.ts} +1 -1
- package/dist/evaluation/axis-checks/index.d.ts +3 -3
- package/dist/evaluation/axis-checks/index.js +46 -0
- package/dist/evaluation/axis-checks/index.js.map +1 -1
- package/dist/evaluation/axis-checks/registry.d.ts +1 -1
- package/dist/evaluation/axis-checks/registry.js +46 -0
- package/dist/evaluation/axis-checks/registry.js.map +1 -1
- package/dist/evaluation/index.d.ts +38 -6
- package/dist/evaluation/index.js +682 -3357
- package/dist/evaluation/index.js.map +1 -1
- package/dist/harness/check/runtime-render/index.d.ts +2 -2
- package/dist/harness/check/runtime-render/index.js +13 -3
- package/dist/harness/check/runtime-render/index.js.map +1 -1
- package/dist/harness/check/runtime-render/render-check-worker.js +8 -3
- package/dist/harness/check/runtime-render/render-check-worker.js.map +1 -1
- package/dist/harness/index.d.ts +2 -2
- package/dist/harness/index.js +562 -30
- package/dist/harness/index.js.map +1 -1
- package/dist/harness/llm-trace-sink.d.ts +7 -1
- package/dist/harness/llm-trace-sink.js.map +1 -1
- package/dist/harness/result-types.d.ts +1 -1
- package/dist/harness/runtime.d.ts +11 -2
- package/dist/harness/runtime.js +3 -3
- package/dist/harness/runtime.js.map +1 -1
- package/dist/harness/types-public.d.ts +1 -1
- package/dist/harness/types-public.js +2 -2
- package/dist/harness/types-public.js.map +1 -1
- package/dist/harness/validator-trace-sink.d.ts +1 -1
- package/dist/index.d.ts +3 -3
- package/dist/index.js +1060 -3520
- package/dist/index.js.map +1 -1
- package/dist/{llm-router-D7fjBhtr.d.ts → llm-router-DwwUK1bz.d.ts} +16 -3
- package/dist/patch.d.ts +1 -1
- package/dist/providers/index.js +5 -4
- package/dist/providers/index.js.map +1 -1
- package/dist/{render-check-WfOVO9kU.d.ts → render-check-C0CqVceF.d.ts} +17 -0
- package/dist/{types-public-DKR_h70H.d.ts → types-public-CzYqZclY.d.ts} +44 -1
- package/dist/validation/index.d.ts +1 -1
- package/dist/validation/index.js +1 -1
- package/dist/validation/index.js.map +1 -1
- package/dist/workflows.d.ts +1 -1
- package/package.json +16 -12
- package/src/adapters/base.ts +8 -0
- package/src/adapters/claude/claude-code-login.ts +73 -0
- package/src/adapters/claude/sdk.ts +8 -1
- package/src/adapters/generation-dispatch.ts +19 -2
- package/src/adapters/openai/raw.ts +11 -5
- package/src/adapters/openai/sdk.ts +8 -5
- package/src/adapters/openai/tokens.ts +67 -0
- package/src/adapters/provider-router.ts +28 -0
- package/src/adapters/types.ts +1 -1
- package/src/boilerplate/generate.ts +19 -4
- package/src/boilerplate/hard-sections.ts +1 -1
- package/src/create-ui-generator.ts +19 -10
- package/src/evaluation/axis-checks/checks/universal.ts +92 -0
- package/src/evaluation/design-identity.ts +16 -5
- package/src/evaluation/types-public.ts +42 -0
- package/src/evaluation/visual-evaluator.ts +231 -24
- package/src/harness/check/runtime-render/adapter.ts +9 -0
- package/src/harness/check/runtime-render/render-check.ts +37 -1
- package/src/harness/claude-code-login-query.ts +528 -0
- package/src/harness/coding/assemble-result.ts +10 -1
- package/src/harness/coding/run-eval-round.ts +179 -34
- package/src/harness/enforced-coding.ts +20 -0
- package/src/harness/llm-router.ts +390 -55
- package/src/harness/llm-trace-sink.ts +7 -1
- package/src/harness/runtime.ts +6 -0
- package/src/providers/openai.ts +31 -19
- package/src/validation/primitives.ts +1 -1
|
@@ -83,6 +83,23 @@ interface CriterionCoverage {
|
|
|
83
83
|
/** Populated for `skipped` / `not-applicable` — why no verdict. */
|
|
84
84
|
readonly reason?: string;
|
|
85
85
|
}
|
|
86
|
+
/**
|
|
87
|
+
* Coverage of the screenshot JUDGE leg for one eval round (ggui#1221) — a
|
|
88
|
+
* different surface from the `visual` CRITERION in `criteriaCoverage` (the
|
|
89
|
+
* LLM evaluator's read of the source). `ran` = a frame was captured and
|
|
90
|
+
* judged; `skipped` = the leg was configured but produced no verdict
|
|
91
|
+
* (no browser, a launch or in-page failure, an unbundlable component, an
|
|
92
|
+
* unparsable judge answer) with `reason` naming why; `not-applicable` =
|
|
93
|
+
* the round had no visual leg configured. Before this row a skipped leg
|
|
94
|
+
* left NO trace on the result — `{ issues: [] }` read exactly like a
|
|
95
|
+
* clean run, and every bootstrap mint on a browserless runtime "passed"
|
|
96
|
+
* its visual round that way.
|
|
97
|
+
*/
|
|
98
|
+
interface VisualCoverage {
|
|
99
|
+
readonly status: CriterionRunStatus;
|
|
100
|
+
/** Populated for `skipped` / `not-applicable` — why no verdict. */
|
|
101
|
+
readonly reason?: string;
|
|
102
|
+
}
|
|
86
103
|
/**
|
|
87
104
|
* The static criteria `runLLMEvaluation` runs, in stamp order — the
|
|
88
105
|
* ONE list both the evaluator's coverage stamp and the harness's
|
|
@@ -131,6 +148,13 @@ interface EvalResult {
|
|
|
131
148
|
* never here — this object is serialized into reports.
|
|
132
149
|
*/
|
|
133
150
|
visual?: VisualEvalSummary;
|
|
151
|
+
/**
|
|
152
|
+
* Whether the screenshot judge leg ran, was skipped (with the reason) or
|
|
153
|
+
* was not configured — ALWAYS stamped by the harness's eval round
|
|
154
|
+
* (ggui#1221), so a consumer can tell a clean visual run from a leg
|
|
155
|
+
* that silently could not run. See {@link VisualCoverage}.
|
|
156
|
+
*/
|
|
157
|
+
visualCoverage?: VisualCoverage;
|
|
134
158
|
}
|
|
135
159
|
/** One canvas's visual verdict, PNG-free — see `EvalResult.visual`. */
|
|
136
160
|
/**
|
|
@@ -173,6 +197,23 @@ interface CanvasVisualSummary {
|
|
|
173
197
|
*/
|
|
174
198
|
fit?: 'fill';
|
|
175
199
|
}
|
|
200
|
+
/**
|
|
201
|
+
* ggui#1195 — the fit reading on the canvas whose policy FAILS an overflow
|
|
202
|
+
* (the inline chat card): the box it was judged at (`ceiling` — the order's
|
|
203
|
+
* declared viewport when one was carried, else the class box), whether that
|
|
204
|
+
* box was declared, and how many px of content sit below it (0 = fits).
|
|
205
|
+
* Present only when such a canvas was judged with a measurable height, so
|
|
206
|
+
* the bar and the ledger read one number for "does the hello fit".
|
|
207
|
+
*/
|
|
208
|
+
interface VisualFitStamp {
|
|
209
|
+
readonly canvas: CanvasClass;
|
|
210
|
+
readonly ceiling: {
|
|
211
|
+
readonly width: number;
|
|
212
|
+
readonly height: number;
|
|
213
|
+
};
|
|
214
|
+
readonly declared: boolean;
|
|
215
|
+
readonly overflowPx: number;
|
|
216
|
+
}
|
|
176
217
|
/** The visual leg's per-canvas summary — see `EvalResult.visual`. */
|
|
177
218
|
interface VisualEvalSummary {
|
|
178
219
|
/** Mean of the canvas scores (rounded). */
|
|
@@ -180,6 +221,8 @@ interface VisualEvalSummary {
|
|
|
180
221
|
/** Every canvas passed. */
|
|
181
222
|
passed: boolean;
|
|
182
223
|
canvases: CanvasVisualSummary[];
|
|
224
|
+
/** ggui#1195 — see {@link VisualFitStamp}; absent when no fail-policy canvas was judged. */
|
|
225
|
+
fit?: VisualFitStamp;
|
|
183
226
|
/** The design tree the judge painted with (ggui#1042): `src` + `srcSha256` — `design@judge`. */
|
|
184
227
|
design?: {
|
|
185
228
|
readonly src: string;
|
|
@@ -301,4 +344,4 @@ declare function getLLMCriteria(designMode?: DesignMode): EvalCriterion[];
|
|
|
301
344
|
*/
|
|
302
345
|
declare function buildCodingCriteriaSummary(designMode?: DesignMode): string;
|
|
303
346
|
|
|
304
|
-
export { type AxisCheck as A, type CanvasJudgeRecord as C, DEFAULT_QUALITY_CONFIG as D, type EvalIssue as E, LLM_EVAL_STATIC_CRITERIA as L, type Priority as P, type QualityConfig as Q, type RuntimeProbeStatus as R, type VisualEvalSummary as V, type AxisCheckInput as a, type EvalResult as b, type EvalTier as c,
|
|
347
|
+
export { type AxisCheck as A, type CanvasJudgeRecord as C, DEFAULT_QUALITY_CONFIG as D, type EvalIssue as E, LLM_EVAL_STATIC_CRITERIA as L, type Priority as P, type QualityConfig as Q, type RuntimeProbeStatus as R, type VisualEvalSummary as V, type AxisCheckInput as a, type EvalResult as b, type EvalTier as c, type VisualCoverage as d, CRITERIA as e, type CanvasVisualSummary as f, type CriterionCoverage as g, type CriterionRunStatus as h, type EvalCategory as i, type EvalCriterion as j, type EvalOutcome as k, type QualityMode as l, matches as m, buildCodingCriteriaSummary as n, criteriaFor as o, getActionableIssues as p, getCriteriaByPriority as q, getCriterionById as r, getLLMCriteria as s, isBlocked as t, notApplicableCoverage as u, priorityForIssue as v };
|