@sanity/ailf 4.0.1 → 4.0.3

package/dist/_vendor/ailf-core/schemas/pipeline-request.d.ts

@@ -76,6 +76,13 @@ export declare const PipelineRequestSchema: z.ZodObject<{
         "content-lake": "content-lake";
         inline: "inline";
     }>>;
+    taskSource: z.ZodOptional<z.ZodObject<{
+        type: z.ZodOptional<z.ZodEnum<{
+            "content-lake": "content-lake";
+            repo: "repo";
+        }>>;
+        repoTasksPath: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>;
     tasks: z.ZodOptional<z.ZodArray<z.ZodString>>;
     urls: z.ZodOptional<z.ZodArray<z.ZodString>>;
     variant: z.ZodOptional<z.ZodEnum<{

package/dist/_vendor/ailf-core/schemas/pipeline-request.js

@@ -126,6 +126,24 @@ export const PipelineRequestSchema = z.object({
     source: z.string().optional(),
     sourceReportId: z.string().optional(),
     taskMode: z.enum(["content-lake", "inline"]).optional(),
+    /**
+     * Task-source configuration (W0077 Phase 6h). Mirrors
+     * `EvalConfigSchema.taskSource` so consumers can express task-source
+     * intent through the universal request payload instead of CLI flags.
+     *
+     * - `type`           — `content-lake` (default) or `repo`
+     * - `repoTasksPath`  — path interpreted relative to `rootDir` when
+     *                      mapped through `mapRequestToConfig`
+     *
+     * When both `taskSource.type` and the legacy `taskMode` are present,
+     * `taskSource.type` wins.
+     */
+    taskSource: z
+        .object({
+        type: z.enum(["content-lake", "repo"]).optional(),
+        repoTasksPath: z.string().min(1).optional(),
+    })
+        .optional(),
     tasks: z.array(z.string()).optional(),
     urls: z.array(z.string().url()).optional(),
     /**

package/dist/_vendor/ailf-core/types/index.d.ts

@@ -1224,8 +1224,11 @@ export interface ComparisonReport {
     deltas: {
         /** Overall score delta (experiment.avgScore − baseline.avgScore) */
         overall: number;
-        /** Per-area total score deltas */
-        perArea: Record<string, number>;
+        /** Per-area total score deltas (array shape — W0137 / D0041) */
+        perArea: {
+            area: string;
+            delta: number;
+        }[];
         /** Per-dimension average deltas (across all areas) */
         perDimension: Record<string, number>;
         /** Doc Lift average delta */

package/dist/adapters/api-client/build-request.js

@@ -178,44 +178,17 @@ export function resolveTasksDir(rootDir, explicitPath) {
 // Helpers
 // ---------------------------------------------------------------------------
 /**
- * Convert a GeneralizedTaskDefinition to the camelCase inline format expected
- * by the API.
+ * Convert a `GeneralizedTaskDefinition` to the wire format expected by the
+ * pipeline's `inlineTasks` array. The canonical shape *is* the wire format —
+ * `CanonicalTaskSchema` (the receiving validator in
+ * `packages/eval/src/adapters/task-sources/repo-schemas.ts`) mirrors this
+ * type per-variant via `.strict()`, so any reshaping here would either be a
+ * no-op or rejected on the receiver. The helper exists as a typed boundary
+ * between the typed task array and `PipelineRequestSchema.inlineTasks`'s
+ * `Record<string, unknown>[]` shape.
  */
 function taskToInlineFormat(task) {
-    const inline = {
-        id: task.id,
-        mode: task.mode,
-        description: task.title,
-        featureArea: task.area ?? "",
-        assert: task.assertions ?? [],
-    };
-    if (task.context?.docs?.length) {
-        inline.canonicalDocs = task.context.docs;
-    }
-    const taskPrompt = task.prompt?.text ?? "";
-    if (taskPrompt) {
-        inline.vars = {
-            task: taskPrompt,
-            docs: "",
-            ...(task.prompt?.vars ?? {}),
-        };
-    }
-    // Literacy-specific fields
-    if (task.mode === "literacy") {
-        if (task.docCoverage) {
-            inline.docCoverage = true;
-        }
-        if (task.referenceSolution) {
-            inline.referenceSolution = task.referenceSolution;
-        }
-        if (task.baseline) {
-            inline.baseline = task.baseline;
-        }
-    }
-    if (task.tags?.length) {
-        inline.tags = task.tags;
-    }
-    return inline;
+    return task;
 }
 /**
  * Build a descriptive error when the task list is empty after filtering.

package/dist/adapters/api-client/format-error.d.ts

@@ -16,7 +16,7 @@ import type { JobResponse } from "./types.js";
  * ❌ Pipeline failed at step 'fetch-docs'
  *    Postcondition failed: Canonical context for task "foo" is empty.
  *
- * 💡 One or more canonicalDocs slugs in your task definitions don't match ...
+ * 💡 One or more context.docs entries in your task definitions don't resolve ...
  * ```
  */
 export declare function formatJobError(job: JobResponse): string;

package/dist/adapters/api-client/format-error.js

@@ -16,7 +16,7 @@ import { getRemediationHint } from "./remediation.js";
  * ❌ Pipeline failed at step 'fetch-docs'
  *    Postcondition failed: Canonical context for task "foo" is empty.
  *
- * 💡 One or more canonicalDocs slugs in your task definitions don't match ...
+ * 💡 One or more context.docs entries in your task definitions don't resolve ...
  * ```
  */
 export function formatJobError(job) {

package/dist/adapters/api-client/remediation.js

@@ -15,9 +15,10 @@ const HINTS = [
     {
         match: (e) => /canonical context.*empty/i.test(e.message) ||
             /no article found for slug/i.test(e.message),
-        hint: "One or more `canonicalDocs` slugs in your task definitions don't match " +
-            "any article in the documentation. Check the `slug` values in " +
-            "`.ailf/tasks/` and ensure they correspond to real articles.\n" +
+        hint: "One or more `context.docs` entries in your task definitions don't " +
+            "resolve to any article in the documentation. Check the `slug`, " +
+            "`path`, or `id` values in `.ailf/tasks/` and ensure they correspond " +
+            "to real articles.\n" +
             "  Run `ailf validate` to check your task definitions locally.",
     },
     {
@@ -49,9 +50,9 @@ const HINTS = [
     {
         match: (e) => e.step === "fetch-docs" && /postcondition/i.test(e.message),
         hint: "The documentation fetch step completed but one or more tasks had " +
-            "empty context. This usually means a `canonicalDocs` slug doesn't " +
-            "match any article.\n" +
-            "  Check the slug values in `.ailf/tasks/`.",
+            "empty context. This usually means a `context.docs` entry doesn't " +
+            "resolve to any article.\n" +
+            "  Check the slug/path/id values in `.ailf/tasks/`.",
     },
     {
         match: (e) => e.step === "dispatch" && /dispatch failed/i.test(e.message),

package/dist/adapters/task-sources/index.d.ts

@@ -1,6 +1,6 @@
 export { CompositeTaskSource } from "./composite-task-source.js";
 export { ContentLakeTaskSource } from "./content-lake-task-source.js";
-export { CanonicalTaskFileSchema, CanonicalTaskSchema, ContentLakeAuthorableTaskSchema, CURATED_ASSERTION_TYPES, detectLegacyFieldNames, parseCanonicalTaskFile, parseRepoConfig, RepoConfigSchema, RUBRIC_TEMPLATE_NAMES, type CanonicalTask, type ContentLakeAuthorableTaskParsed, type CuratedAssertionType, type RepoConfig, type RubricTemplateName, } from "./repo-schemas.js";
+export { AilfEvalWorkflowSchema, CanonicalTaskFileSchema, CanonicalTaskSchema, ContentLakeAuthorableTaskSchema, CURATED_ASSERTION_TYPES, detectLegacyFieldNames, parseAilfEvalWorkflow, parseCanonicalTaskFile, parseRepoConfig, RepoConfigSchema, RUBRIC_TEMPLATE_NAMES, type AilfEvalWorkflow, type CanonicalTask, type ContentLakeAuthorableTaskParsed, type CuratedAssertionType, type RepoConfig, type RubricTemplateName, } from "./repo-schemas.js";
 export { RepoTaskSource } from "./repo-task-source.js";
 export { detectTriggerContext, resolveTrigger, type ResolvedTrigger, type TriggerContext, } from "./repo-trigger.js";
 export { formatValidationResult, validateCanonicalTasks, type ValidationMessage, type ValidationResult, } from "./repo-validation.js";

package/dist/adapters/task-sources/index.js

@@ -1,6 +1,6 @@
 export { CompositeTaskSource } from "./composite-task-source.js";
 export { ContentLakeTaskSource } from "./content-lake-task-source.js";
-export { CanonicalTaskFileSchema, CanonicalTaskSchema, ContentLakeAuthorableTaskSchema, CURATED_ASSERTION_TYPES, detectLegacyFieldNames, parseCanonicalTaskFile, parseRepoConfig, RepoConfigSchema, RUBRIC_TEMPLATE_NAMES, } from "./repo-schemas.js";
+export { AilfEvalWorkflowSchema, CanonicalTaskFileSchema, CanonicalTaskSchema, ContentLakeAuthorableTaskSchema, CURATED_ASSERTION_TYPES, detectLegacyFieldNames, parseAilfEvalWorkflow, parseCanonicalTaskFile, parseRepoConfig, RepoConfigSchema, RUBRIC_TEMPLATE_NAMES, } from "./repo-schemas.js";
 export { RepoTaskSource } from "./repo-task-source.js";
 export { detectTriggerContext, resolveTrigger, } from "./repo-trigger.js";
 export { formatValidationResult, validateCanonicalTasks, } from "./repo-validation.js";

package/dist/adapters/task-sources/repo-schemas.d.ts

@@ -1526,3 +1526,34 @@ export type RepoConfig = z.infer<typeof RepoConfigSchema>;
  * Parse and validate .ailf/config.yaml content. Returns typed config or throws.
  */
 export declare function parseRepoConfig(raw: unknown, filename?: string): RepoConfig;
+/**
+ * Structural schema for the `ailf-eval.yml` workflow template emitted by
+ * `ailf init`. Validates the consumer-visible contract: a workflow named
+ * "AI Literacy Eval" with at least one job containing checkout + eval
+ * steps. The literal YAML body is intentionally not pinned — comments,
+ * step ordering, and option flags can shift without breaking consumers.
+ */
+export declare const AilfEvalWorkflowSchema: z.ZodObject<{
+    name: z.ZodString;
+    on: z.ZodUnknown;
+    jobs: z.ZodRecord<z.ZodString, z.ZodObject<{
+        name: z.ZodOptional<z.ZodString>;
+        "runs-on": z.ZodUnion<readonly [z.ZodString, z.ZodArray<z.ZodString>]>;
+        permissions: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+        steps: z.ZodArray<z.ZodObject<{
+            name: z.ZodOptional<z.ZodString>;
+            id: z.ZodOptional<z.ZodString>;
+            if: z.ZodOptional<z.ZodString>;
+            uses: z.ZodOptional<z.ZodString>;
+            run: z.ZodOptional<z.ZodString>;
+            env: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+            with: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+        }, z.core.$loose>>;
+    }, z.core.$loose>>;
+}, z.core.$loose>;
+export type AilfEvalWorkflow = z.infer<typeof AilfEvalWorkflowSchema>;
+/**
+ * Parse and validate a `.github/workflows/ailf-eval.yml` payload (already
+ * loaded from YAML). Throws with a Zod-formatted message on failure.
+ */
+export declare function parseAilfEvalWorkflow(raw: unknown, filename?: string): AilfEvalWorkflow;

package/dist/adapters/task-sources/repo-schemas.js

@@ -602,3 +602,61 @@ export function parseRepoConfig(raw, filename = ".ailf/config.yaml") {
     }
     return result.data;
 }
+// ---------------------------------------------------------------------------
+// Workflow contract — structural shape of `.github/workflows/ailf-eval.yml`
+//
+// Used by the Tier 1 init contract test (W0139 M2) and the Tier 4 drift
+// check to assert that the template `init` writes still matches what a
+// downstream consumer's PR comment + report-publish path depends on.
+//
+// Loose by design: this is a contract on the consumer-visible parts (job
+// runs `npx @sanity/ailf ... run --remote`, posts a PR comment, names the
+// workflow `AI Literacy Eval`), not the literal byte content of the YAML.
+// ---------------------------------------------------------------------------
+const WorkflowStepSchema = z
+    .object({
+    name: z.string().optional(),
+    id: z.string().optional(),
+    if: z.string().optional(),
+    uses: z.string().optional(),
+    run: z.string().optional(),
+    env: z.record(z.string(), z.unknown()).optional(),
+    with: z.record(z.string(), z.unknown()).optional(),
+})
+    .passthrough();
+const WorkflowJobSchema = z
+    .object({
+    name: z.string().optional(),
+    "runs-on": z.union([z.string(), z.array(z.string())]),
+    permissions: z.record(z.string(), z.unknown()).optional(),
+    steps: z.array(WorkflowStepSchema).min(1),
+})
+    .passthrough();
+/**
+ * Structural schema for the `ailf-eval.yml` workflow template emitted by
+ * `ailf init`. Validates the consumer-visible contract: a workflow named
+ * "AI Literacy Eval" with at least one job containing checkout + eval
+ * steps. The literal YAML body is intentionally not pinned — comments,
+ * step ordering, and option flags can shift without breaking consumers.
+ */
+export const AilfEvalWorkflowSchema = z
+    .object({
+    name: z.string().min(1),
+    on: z.unknown(),
+    jobs: z.record(z.string(), WorkflowJobSchema),
+})
+    .passthrough();
+/**
+ * Parse and validate a `.github/workflows/ailf-eval.yml` payload (already
+ * loaded from YAML). Throws with a Zod-formatted message on failure.
+ */
+export function parseAilfEvalWorkflow(raw, filename = ".github/workflows/ailf-eval.yml") {
+    const result = AilfEvalWorkflowSchema.safeParse(raw);
+    if (!result.success) {
+        const messages = result.error.issues
+            .map((i) => `  [${i.path.join(".")}]: ${i.message}`)
+            .join("\n");
+        throw new Error(`Invalid workflow "${filename}":\n${messages}`);
+    }
+    return result.data;
+}

package/dist/commands/init.d.ts

@@ -17,4 +17,11 @@
  *   ailf init --path ./my-dir        # target a specific directory
  */
 import { Command } from "commander";
+export interface InitOptions {
+    format: "ts" | "yaml" | "json";
+    force: boolean;
+    path: string;
+    mode?: string;
+}
 export declare function createInitCommand(): Command;
+export declare function runInit(opts: InitOptions): Promise<void>;

package/dist/commands/init.js

@@ -61,7 +61,7 @@ function taskStemsForMode(mode) {
 // ---------------------------------------------------------------------------
 // Init logic
 // ---------------------------------------------------------------------------
-async function runInit(opts) {
+export async function runInit(opts) {
     const validFormats = new Set(["ts", "yaml", "json"]);
     if (!validFormats.has(opts.format)) {
         console.error(`  ✗ Invalid output format "${opts.format}". Valid options: ts, yaml, json`);

package/dist/pipeline/compare.js

@@ -71,11 +71,13 @@ export function compare(baseline, experiment, options) {
     const notEvaluated = areas
         .filter((a) => a.change === "not-evaluated")
         .map((a) => a.area);
-    // Per-area deltas as a record
-    const perArea = {};
-    for (const a of areas) {
-        perArea[a.area] = a.delta;
-    }
+    // Per-area deltas as an array (W0137 / D0041) — keyed-map shapes are
+    // capped by the Sanity attribute limit because each new feature area
+    // mints a permanent attribute path.
+    const perArea = areas.map((a) => ({
+        area: a.area,
+        delta: a.delta,
+    }));
     // Per-dimension average deltas (only for areas present in both summaries)
     const commonAreas = areas.filter((a) => baselineAreas.has(a.area) && experimentAreas.has(a.area));
     const commonCount = commonAreas.length || 1;

package/dist/pipeline/map-request-to-config.js

@@ -61,14 +61,16 @@ export function mapRequestToConfig(request, rootDir) {
         datasetOverride: request.dataset,
         projectIdOverride: request.projectId,
         perspectiveOverride: request.perspective,
-        taskSourceType: mapTaskSourceType(request.taskMode),
+        taskSourceType: mapTaskSourceType(request.taskSource?.type, request.taskMode),
         outputPath: undefined,
         promptfooUrl: undefined,
         studioOriginOverride: undefined,
         sanityDocumentArgs: undefined,
         sourceReportId: request.sourceReportId,
         beforeOption: undefined,
-        repoTasksPath: undefined,
+        repoTasksPath: request.taskSource?.repoTasksPath
+            ? resolve(rootDir, request.taskSource.repoTasksPath)
+            : undefined,
         callerGit: request.callerGit,
         callerEnvelope: buildCallerEnvelope(request),
         callback: request.callback,
@@ -107,14 +109,19 @@ function buildCallerEnvelope(request) {
     }
     return { classification, owner, executor, purpose, labels };
 }
-function mapTaskSourceType(taskMode) {
+function mapTaskSourceType(taskSourceType, taskMode) {
+    // `taskSource.type` is the canonical field; honor it first when set.
+    if (taskSourceType === "repo")
+        return "repo";
+    if (taskSourceType === "content-lake")
+        return "content-lake";
+    // Fall back to the legacy `taskMode` field.
     if (taskMode === "content-lake")
-        return taskMode;
+        return "content-lake";
     // "inline" means the caller sent inline tasks that will be materialized
-    // to a temp directory and loaded via --repo-tasks-path. Use "repo" to
-    // ensure ONLY those tasks are used (no Content Lake merge).
+    // to a temp directory. Use "repo" to ensure ONLY those tasks are used
+    // (no Content Lake merge).
     if (taskMode === "inline")
         return "repo";
-    // "yaml" was removed — treat it as default (Content Lake)
     return undefined;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity/ailf",
-  "version": "4.0.1",
+  "version": "4.0.3",
   "private": false,
   "publishConfig": {
     "access": "public"
@@ -77,7 +77,8 @@
     "test": "tsx --test src/__tests__/*.test.ts src/adapters/**/__tests__/*.adapter.test.ts",
     "test:e2e": "AILF_E2E=1 tsx --test src/__tests__/e2e/*.e2e.test.ts",
     "test:e2e:adapters": "AILF_E2E=1 tsx --test src/adapters/**/__tests__/*.adapter.test.ts",
-    "test:e2e:api": "AILF_E2E_API=1 tsx --test src/__tests__/api-tier2-tenant-integration.test.ts",
+    "test:e2e:api": "AILF_E2E_API=1 tsx --test src/__tests__/api-tier2-tenant-integration.test.ts src/__tests__/gcs-artifact-writer-roundtrip.test.ts",
+    "test:tier3:roundtrip": "AILF_E2E_API=1 AILF_E2E_GITHUB_DISPATCH=1 tsx --test src/__tests__/api-tier3-round-trip.test.ts",
     "test:all": "AILF_E2E=1 tsx --test src/__tests__/*.test.ts src/pipeline/compiler/__tests__/*.test.ts src/__tests__/e2e/*.e2e.test.ts src/adapters/**/__tests__/*.adapter.test.ts",
     "pr-comment": "tsx src/cli.ts pr-comment",
     "coverage-audit": "tsx src/cli.ts report coverage",