@sanity/ailf 2.0.0 → 2.0.2

package/dist/pipeline/grader-validation.js

@@ -11,7 +11,7 @@
  *
  * This module has NO side effects — no file I/O, no API calls.
  *
- * @see docs/exec-plans/grader-reliability.md — Phase 2
+ * @see docs/archive/exec-plans/grader-reliability.md — Phase 2
  */
 // ---------------------------------------------------------------------------
 // Pure computation

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

@@ -1,3 +1,4 @@
+import { resolve } from "node:path";
 import { normalizeMode } from "./normalize-mode.js";
 /**
  * Map a PipelineRequest to a ResolvedConfig.
@@ -19,13 +20,20 @@ import { normalizeMode } from "./normalize-mode.js";
 export function mapRequestToConfig(request, rootDir) {
     // Normalize mode so downstream pipeline code only sees canonical names.
     // The API may receive legacy names ("baseline", "full") from older clients.
-    const { mode, variant } = normalizeMode(request.mode ?? "full");
+    const { mode, variant: normalizedVariant } = normalizeMode(request.mode ?? "full");
+    // Explicit variant from request takes precedence over one derived from
+    // legacy mode normalization. This supports the canonical form:
+    //   { mode: "literacy", variant: "baseline" }
+    // while preserving backward compatibility with:
+    //   { mode: "baseline" } → normalizeMode → { mode: "literacy", variant: "baseline" }
+    const variant = request.variant ?? normalizedVariant;
     // API-triggered evaluations (identified by jobId) default to publish: true.
     // Without this, the job's reportId is always null and GET /v1/reports/:id
     // has nothing to return.
     const publishDefault = !!request.jobId;
     return {
         rootDir,
+        outputDir: resolve(rootDir, "results", "latest"),
         mode,
         variant,
         debug: mapDebug(request.debug),
@@ -66,6 +74,10 @@ export function mapRequestToConfig(request, rootDir) {
         callerGit: request.callerGit,
         callback: request.callback,
         jobId: request.jobId,
+        captureEnabled: false,
+        captureDir: undefined,
+        captureCompress: true,
+        captureExtras: true,
         remote: false,
         apiUrl: "https://ailf-api.sanity.build",
         presets: request.presets,
@@ -84,12 +96,13 @@ function mapDebug(debug) {
     };
 }
 function mapTaskSourceType(taskMode) {
-    if (taskMode === "content-lake" || taskMode === "yaml")
+    if (taskMode === "content-lake")
         return taskMode;
     // "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).
     if (taskMode === "inline")
         return "repo";
+    // "yaml" was removed — treat it as default (Content Lake)
     return undefined;
 }

package/dist/pipeline/mirror-repo-tasks.d.ts

@@ -10,7 +10,7 @@
  * means unchanged tasks are skipped. Changed tasks are upserted via
  * createOrReplace.
  *
- * @see docs/exec-plans/tasks-as-content/phase-5-content-lake-mirroring.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-5-content-lake-mirroring.md
  */
 import type { SanityClient } from "@sanity/client";
 import { type LiteracyTaskDefinition, type Logger } from "../_vendor/ailf-core/index.d.ts";

package/dist/pipeline/mirror-repo-tasks.js

@@ -10,7 +10,7 @@
  * means unchanged tasks are skipped. Changed tasks are upserted via
  * createOrReplace.
  *
- * @see docs/exec-plans/tasks-as-content/phase-5-content-lake-mirroring.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-5-content-lake-mirroring.md
  */
 import { createHash } from "crypto";
 import { readFileSync } from "fs";

package/dist/pipeline/plan-format.d.ts

@@ -7,7 +7,7 @@
  * emoji markers, alignment, and color-coding (via unicode markers).
  * The JSON formatter produces machine-readable output for CI/CD.
  *
- * @see docs/exec-plans/execution-preview.md
+ * @see docs/archive/exec-plans/execution-preview.md
  */
 import type { ExecutionPlan } from "./plan.js";
 /**

package/dist/pipeline/plan-format.js

@@ -7,7 +7,7 @@
  * emoji markers, alignment, and color-coding (via unicode markers).
  * The JSON formatter produces machine-readable output for CI/CD.
  *
- * @see docs/exec-plans/execution-preview.md
+ * @see docs/archive/exec-plans/execution-preview.md
  */
 import { formatCost } from "../agent-observer/pricing.js";
 // ---------------------------------------------------------------------------

package/dist/pipeline/plan.d.ts

@@ -7,7 +7,7 @@
  * anything. Calls existing pure functions (task expansion, model loading,
  * cache hashing, pricing) and composes them into an `ExecutionPlan`.
  *
- * @see docs/exec-plans/execution-preview.md
+ * @see docs/archive/exec-plans/execution-preview.md
  */
 import type { DebugOptions, EvalMode } from "./types.js";
 import { LiteracyVariant } from "./normalize-mode.js";

package/dist/pipeline/plan.js

@@ -7,16 +7,17 @@
  * anything. Calls existing pure functions (task expansion, model loading,
  * cache hashing, pricing) and composes them into an `ExecutionPlan`.
  *
- * @see docs/exec-plans/execution-preview.md
+ * @see docs/archive/exec-plans/execution-preview.md
  */
 import { existsSync, readdirSync, statSync } from "fs";
 import { resolve } from "path";
+import { createLiteracyModeBase, modelMatchesLiteracyVariant, } from "./compiler/mode-bases/literacy.js";
 import { lookupPricing } from "../agent-observer/pricing.js";
 import { RepoTaskSource } from "../adapters/task-sources/repo-task-source.js";
 import { loadAllTsTaskFiles } from "../adapters/task-sources/task-file-loader.js";
 import { lookupCache } from "./cache.js";
 import { compileLiteracyTasks } from "./compiler/literacy-bridge.js";
-import { tryLoadConfigFile } from "./compiler/config-loader.js";
+import { resolveVendoredSubdir, tryLoadConfigFile, } from "./compiler/config-loader.js";
 import { LiteracyVariant } from "./normalize-mode.js";
 import { validateConfiguration } from "./validate.js";
 /**
@@ -44,33 +45,35 @@ function loadModelsFile(rootDir) {
     const result = tryLoadConfigFile("models", rootDir);
     return result?.data ?? null;
 }
+const _literacyBase = createLiteracyModeBase();
 /**
- * Map eval mode + variant to the model "modes" array values from models config.
+ * Check whether a model participates in a given eval mode + optional variant.
  *
- * Literacy mode uses the variant to determine which model sub-modes match.
- * Non-literacy modes accept all models by default (filtering is done
- * elsewhere for those modes).
+ * For literacy mode, checks both mode enrollment and variant participation
+ * via the shared `modelMatchesLiteracyVariant` helper. For non-literacy
+ * modes, checks mode enrollment only.
  */
-function modeMatchesModelModes(mode, modelModes, variant) {
-    if (!modelModes || modelModes.length === 0)
-        return true;
-    if (mode === "literacy") {
+function modeMatchesModel(mode, model, variant) {
+    // Check basic mode enrollment
+    if (model.modes &&
+        model.modes.length > 0 &&
+        !model.modes.includes(mode)) {
+        return false;
+    }
+    // For literacy mode with a variant, check variant participation
+    if (mode === "literacy" && variant) {
         switch (variant) {
             case LiteracyVariant.AGENTIC:
-                return (modelModes.includes("agentic-naive") ||
-                    modelModes.includes("agentic-optimized"));
-            case LiteracyVariant.OBSERVED:
-                return modelModes.includes(LiteracyVariant.OBSERVED);
+                return (modelMatchesLiteracyVariant(model, "agentic-naive") ||
+                    modelMatchesLiteracyVariant(model, "agentic-optimized"));
             case LiteracyVariant.FULL:
-                return (modelModes.includes(LiteracyVariant.STANDARD) ||
-                    modelModes.includes("agentic-naive") ||
-                    modelModes.includes("agentic-optimized"));
-            case LiteracyVariant.STANDARD:
+                return (modelMatchesLiteracyVariant(model, "baseline") ||
+                    modelMatchesLiteracyVariant(model, "agentic-naive") ||
+                    modelMatchesLiteracyVariant(model, "agentic-optimized"));
             default:
-                return modelModes.includes(LiteracyVariant.STANDARD);
+                return modelMatchesLiteracyVariant(model, variant);
         }
     }
-    // Non-literacy modes accept all models by default
     return true;
 }
 // ---------------------------------------------------------------------------
@@ -139,8 +142,8 @@ export async function buildPipelinePlan(opts, rootDir) {
         const modelsForCompile = loadModelsFile(rootDir);
         const graderProvider = modelsForCompile?.grader?.id ?? "openai:chat:gpt-4o";
         const modelEntries = (modelsForCompile?.models ?? []).map((m) => ({ id: m.id, label: m.label }));
-        // Load *.task.ts files from tasks/<mode>/
-        const modeTasksDir = resolve(rootDir, "tasks", opts.mode);
+        // Load *.task.ts files from tasks/<mode>/ (or dist/tasks/<mode>/ when vendored)
+        const modeTasksDir = resolveVendoredSubdir(rootDir, `tasks/${opts.mode}`);
         if (existsSync(modeTasksDir)) {
             const rawTasks = await loadAllTsTaskFiles(modeTasksDir);
             if (rawTasks.length > 0) {
@@ -148,9 +151,16 @@ export async function buildPipelinePlan(opts, rootDir) {
                 const handlerModulePath = `./compiler/mode-handlers/${opts.mode}/index.js`;
                 const mod = await import(handlerModulePath);
                 const handler = mod.handler;
+                const skippedByMode = new Map();
                 for (const rawFile of rawTasks) {
                     for (const taskDef of rawFile.tasks) {
                         const task = taskDef;
+                        // Filter to matching mode (skip tasks from other modes in same dir)
+                        if ("mode" in task && task.mode !== opts.mode) {
+                            const taskMode = task.mode ?? "unknown";
+                            skippedByMode.set(taskMode, (skippedByMode.get(taskMode) ?? 0) + 1);
+                            continue;
+                        }
                         // Apply area/task/tag filter
                         if (filter) {
                             if (filter.areas?.length &&
@@ -192,6 +202,13 @@ export async function buildPipelinePlan(opts, rootDir) {
                         }
                     }
                 }
+                if (skippedByMode.size > 0) {
+                    const summary = [...skippedByMode.entries()]
+                        .map(([m, n]) => `${n} ${m}`)
+                        .join(", ");
+                    const total = [...skippedByMode.values()].reduce((a, b) => a + b, 0);
+                    warnings.push(`Skipped ${total} task(s) with non-matching mode (${summary}). Current pipeline mode: ${opts.mode}. Run with --mode <mode> to include them.`);
+                }
             }
         }
     }
@@ -203,13 +220,29 @@ export async function buildPipelinePlan(opts, rootDir) {
     if (opts.repoTasksPath) {
         try {
             const repoSource = new RepoTaskSource(opts.repoTasksPath);
-            // Type-narrow to literacy tasks — compileLiteracyTasks accepts LiteracyTaskDefinition[]
-            const repoTasks = (await repoSource.loadTasks(filter)).filter((t) => t.mode === "literacy");
+            const allRepoTasks = await repoSource.loadTasks(filter);
+            // Filter to current mode tasks
+            const repoTasks = allRepoTasks.filter((t) => t.mode === opts.mode);
+            const skippedRepoTasks = allRepoTasks.length - repoTasks.length;
+            if (skippedRepoTasks > 0) {
+                const skippedModes = new Map();
+                for (const t of allRepoTasks) {
+                    if (t.mode !== opts.mode) {
+                        skippedModes.set(t.mode, (skippedModes.get(t.mode) ?? 0) + 1);
+                    }
+                }
+                const summary = [...skippedModes.entries()]
+                    .map(([m, n]) => `${n} ${m}`)
+                    .join(", ");
+                warnings.push(`Skipped ${skippedRepoTasks} repo task(s) with non-matching mode (${summary}). Current pipeline mode: ${opts.mode}. Run with --mode <mode> to include them.`);
+            }
             repoTaskCount = repoTasks.length;
-            if (repoTaskCount > 0) {
+            if (repoTaskCount > 0 && opts.mode === "literacy") {
+                // Literacy-specific compilation for repo tasks (detailed test expansion)
+                const literacyRepoTasks = repoTasks.filter((t) => t.mode === "literacy");
                 const modelsForCompile = loadModelsFile(rootDir);
                 const graderProvider = modelsForCompile?.grader?.id ?? "openai:chat:gpt-4o";
-                const compileResult = compileLiteracyTasks(repoTasks, {
+                const compileResult = compileLiteracyTasks(literacyRepoTasks, {
                     rootDir,
                     evalMode: opts.variant === LiteracyVariant.AGENTIC
                         ? LiteracyVariant.AGENTIC
@@ -231,6 +264,11 @@ export async function buildPipelinePlan(opts, rootDir) {
                     }
                 }
             }
+            else if (repoTaskCount > 0) {
+                // Non-literacy modes: approximate 1 test per task (compilation not
+                // supported for non-literacy repo tasks in the explain preview yet)
+                totalTests += repoTaskCount;
+            }
         }
         catch {
             warnings.push(`Failed to scan repo tasks at ${opts.repoTasksPath} — count may be underestimated`);
@@ -244,19 +282,19 @@ export async function buildPipelinePlan(opts, rootDir) {
     const models = [];
     let graderModelName = "";
     if (modelsFile) {
-        const activeModels = modelsFile.models.filter((m) => modeMatchesModelModes(opts.mode, m.modes, opts.variant));
+        const activeModels = modelsFile.models.filter((m) => modeMatchesModel(opts.mode, m, opts.variant));
         // For agentic mode, each model appears twice (naive + optimized)
         for (const m of activeModels) {
             const modelName = extractModelName(m.id);
             if (opts.variant === LiteracyVariant.AGENTIC) {
-                if (m.modes?.includes("agentic-naive")) {
+                if (modelMatchesLiteracyVariant(m, "agentic-naive")) {
                     models.push({
                         id: m.id,
                         label: `${m.label} (Naive)`,
                         modelName,
                     });
                 }
-                if (m.modes?.includes("agentic-optimized")) {
+                if (modelMatchesLiteracyVariant(m, "agentic-optimized")) {
                     models.push({
                         id: m.id,
                         label: `${m.label} (Optimized)`,

package/dist/pipeline/probe.d.ts

@@ -14,7 +14,7 @@
  * not "are these docs good enough?" The output is always labeled as
  * directional and never displayed on the same scale as scored evaluations.
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 import type { ProbeResult } from "./types.js";
 /** Generic probe prompt template */

package/dist/pipeline/probe.js

@@ -14,7 +14,7 @@
  * not "are these docs good enough?" The output is always labeled as
  * directional and never displayed on the same scale as scored evaluations.
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 // ---------------------------------------------------------------------------
 // Constants

package/dist/pipeline/readiness-report.d.ts

@@ -13,8 +13,8 @@
  *   - generateReadinessReport()  — builds the structured report
  *   - formatReadinessMarkdown()  — renders the report as markdown
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-5-readiness-thresholds.md
- * @see docs/exec-plans/eliminate-lib-layer.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-5-readiness-thresholds.md
+ * @see docs/archive/exec-plans/eliminate-lib-layer.md
  */
 import type { ThresholdConfig } from "./schemas.js";
 import type { GapAnalysisReport, GapEstimate, ScoreSummary, ThresholdEvaluation, ThresholdViolation } from "./types.js";

package/dist/pipeline/readiness-report.js

@@ -13,8 +13,8 @@
  *   - generateReadinessReport()  — builds the structured report
  *   - formatReadinessMarkdown()  — renders the report as markdown
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-5-readiness-thresholds.md
- * @see docs/exec-plans/eliminate-lib-layer.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-5-readiness-thresholds.md
+ * @see docs/archive/exec-plans/eliminate-lib-layer.md
  */
 import { evaluateThresholds } from "./thresholds.js";
 // ---------------------------------------------------------------------------

package/dist/pipeline/release-classification.d.ts

@@ -15,7 +15,7 @@
  * - **not-applicable**: Updated, removed, or unchanged documents (these
  *   follow the standard before/after comparison flow from Phase 2).
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 import type { ClassifiedReleaseDocument, ProductFeature, ReleaseClassification } from "./types.js";
 import type { ReverseMapping } from "./reverse-mapping.js";

package/dist/pipeline/release-classification.js

@@ -15,7 +15,7 @@
  * - **not-applicable**: Updated, removed, or unchanged documents (these
  *   follow the standard before/after comparison flow from Phase 2).
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 // ---------------------------------------------------------------------------
 // Public API

package/dist/pipeline/release-report.d.ts

@@ -10,7 +10,7 @@
  * attribution (2c), and probe results (4b) into the document × area × task
  * impact matrix specified by Scenario 2.4.
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 import type { AttributionReport, ComparisonReport, ProbeResult, ReleaseClassification, ReleaseImpactReport } from "./types.js";
 /**

package/dist/pipeline/release-report.js

@@ -10,7 +10,7 @@
  * attribution (2c), and probe results (4b) into the document × area × task
  * impact matrix specified by Scenario 2.4.
  *
- * @see docs/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
+ * @see docs/archive/exec-plans/scenario-matrix-implementation/phase-4-content-release-integration.md
  */
 // ---------------------------------------------------------------------------
 // Public API

package/dist/pipeline/repo-eval-comment.d.ts

@@ -12,7 +12,7 @@
  * - Clear "what does this mean?" context
  * - skip-ailf bypass instructions
  *
- * @see docs/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
  * @see packages/eval/src/pipeline/repo-threshold-evaluator.ts
  */
 import type { ComparisonReport, ScoreSummary } from "./types.js";

package/dist/pipeline/repo-eval-comment.js

@@ -12,7 +12,7 @@
  * - Clear "what does this mean?" context
  * - skip-ailf bypass instructions
  *
- * @see docs/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
  * @see packages/eval/src/pipeline/repo-threshold-evaluator.ts
  */
 // ---------------------------------------------------------------------------

package/dist/pipeline/repo-threshold-evaluator.d.ts

@@ -10,7 +10,7 @@
  * thresholds are per-area, defined by the AILF team, and drive
  * readiness reports.
  *
- * @see docs/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
  * @see packages/eval/src/adapters/task-sources/repo-schemas.ts
  */
 import type { ScoreSummary } from "./types.js";

package/dist/pipeline/repo-threshold-evaluator.js

@@ -10,7 +10,7 @@
  * thresholds are per-area, defined by the AILF team, and drive
  * readiness reports.
  *
- * @see docs/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
+ * @see docs/archive/exec-plans/tasks-as-content/phase-6-pr-quality-gates.md
  * @see packages/eval/src/adapters/task-sources/repo-schemas.ts
  */
 // ---------------------------------------------------------------------------

package/dist/pipeline/resolve-mappings.d.ts

@@ -1,9 +1,9 @@
 /**
  * pipeline/resolve-mappings.ts
  *
- * Extracts canonical mappings from inline task YAML definitions.
- * Each task file contains canonical_docs and reference_solution fields
- * directly — there is no separate mappings.yaml file.
+ * Extracts canonical mappings from task definitions (*.task.ts files).
+ * Each task contains context.docs and referenceSolution fields
+ * directly — there is no separate mappings file.
  *
  * The output shape matches what downstream consumers expect so
  * fetch-docs, validate, and calculate-scores work without changes.
@@ -24,12 +24,12 @@ export interface ResolvedMappings {
     }>;
 }
 /**
- * Extract inline canonical mappings from task YAML files.
- * Only tasks with both an id and canonical_docs fields are included.
+ * Extract canonical mappings from *.task.ts files in tasks/literacy/.
+ * Only tasks with context.docs and referenceSolution are included.
  */
 export declare function extractInlineMappings(rootDir: string): ResolvedMappings;
 /**
- * Resolve canonical mappings from inline task YAML definitions.
+ * Resolve canonical mappings from task definitions.
  * This is the single source of truth — there is no external mappings file.
  */
 export declare function resolveMappings(rootDir: string): ResolvedMappings;

package/dist/pipeline/resolve-mappings.js

@@ -1,72 +1,72 @@
 /**
  * pipeline/resolve-mappings.ts
  *
- * Extracts canonical mappings from inline task YAML definitions.
- * Each task file contains canonical_docs and reference_solution fields
- * directly — there is no separate mappings.yaml file.
+ * Extracts canonical mappings from task definitions (*.task.ts files).
+ * Each task contains context.docs and referenceSolution fields
+ * directly — there is no separate mappings file.
  *
  * The output shape matches what downstream consumers expect so
  * fetch-docs, validate, and calculate-scores work without changes.
  */
-import { existsSync, readFileSync, readdirSync } from "fs";
-import { resolve } from "path";
-import { load } from "js-yaml";
+import { existsSync } from "fs";
+import { discoverTsTaskFiles, loadTsTaskFileSync, } from "../adapters/task-sources/task-file-loader.js";
+import { resolveVendoredSubdir } from "./compiler/config-loader.js";
 // ---------------------------------------------------------------------------
 // Resolution
 // ---------------------------------------------------------------------------
 /**
- * Extract inline canonical mappings from task YAML files.
- * Only tasks with both an id and canonical_docs fields are included.
+ * Extract canonical mappings from *.task.ts files in tasks/literacy/.
+ * Only tasks with context.docs and referenceSolution are included.
  */
 export function extractInlineMappings(rootDir) {
-    const tasksDir = resolve(rootDir, "tasks");
+    const tasksDir = resolveVendoredSubdir(rootDir, "tasks/literacy");
     const result = { feature_areas: {} };
     if (!existsSync(tasksDir))
         return result;
-    const yamlFiles = readdirSync(tasksDir)
-        .filter((f) => (f.endsWith(".yaml") || f.endsWith(".yml")) && !f.startsWith("."))
-        .sort();
-    for (const file of yamlFiles) {
-        const featureArea = file.replace(/\.(yaml|yml)$/, "");
-        const filePath = resolve(tasksDir, file);
-        const raw = readFileSync(filePath, "utf-8");
-        const parsed = load(raw);
-        if (!Array.isArray(parsed))
-            continue;
-        const tasks = [];
-        for (const entry of parsed) {
-            if (!isInlineTaskWithMappings(entry))
+    const files = discoverTsTaskFiles(tasksDir);
+    for (const file of files) {
+        const loaded = loadTsTaskFileSync(file);
+        for (const task of loaded.tasks) {
+            const t = task;
+            const area = typeof t.area === "string" ? t.area : undefined;
+            const id = typeof t.id === "string" ? t.id : undefined;
+            const title = typeof t.title === "string" ? t.title : "";
+            const referenceSolution = typeof t.referenceSolution === "string" ? t.referenceSolution : "";
+            if (!area || !id)
                 continue;
-            tasks.push({
-                canonical_docs: entry.canonical_docs,
-                description: entry.description,
-                id: entry.id,
-                reference_solution: entry.reference_solution,
+            // Extract docs from context.docs (GeneralizedDocRef[])
+            const context = t.context;
+            const docs = [];
+            if (context?.docs && Array.isArray(context.docs)) {
+                for (const doc of context.docs) {
+                    const d = doc;
+                    if (typeof d.slug === "string") {
+                        docs.push({
+                            slug: d.slug,
+                            reason: typeof d.reason === "string" ? d.reason : "",
+                        });
+                    }
+                }
+            }
+            if (docs.length === 0 || !referenceSolution)
+                continue;
+            if (!result.feature_areas[area]) {
+                result.feature_areas[area] = { tasks: [] };
+            }
+            result.feature_areas[area].tasks.push({
+                canonical_docs: docs,
+                description: title,
+                id,
+                reference_solution: referenceSolution,
             });
         }
-        if (tasks.length > 0) {
-            result.feature_areas[featureArea] = { tasks };
-        }
     }
     return result;
 }
 /**
- * Resolve canonical mappings from inline task YAML definitions.
+ * Resolve canonical mappings from task definitions.
  * This is the single source of truth — there is no external mappings file.
  */
 export function resolveMappings(rootDir) {
     return extractInlineMappings(rootDir);
 }
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-function isInlineTaskWithMappings(entry) {
-    if (typeof entry !== "object" || entry === null)
-        return false;
-    const e = entry;
-    return (typeof e.id === "string" &&
-        typeof e.description === "string" &&
-        Array.isArray(e.canonical_docs) &&
-        e.canonical_docs.length > 0 &&
-        typeof e.reference_solution === "string");
-}

package/dist/pipeline/retrieval-metrics.d.ts

@@ -2,9 +2,9 @@
  * pipeline/retrieval-metrics.ts
  *
  * Computes retrieval precision and recall by comparing agent-retrieved
- * doc slugs against canonical_docs defined in task YAML files.
+ * doc slugs against canonical_docs defined in task definitions.
  *
- * This is a pure computation module — no file I/O beyond reading task YAMLs.
+ * This is a pure computation module — no file I/O beyond reading task files.
  */
 import type { RetrievalMetrics, TaskRetrievalMetrics } from "./types.js";
 export interface AgenticBehaviorData {
@@ -30,7 +30,7 @@ export declare function computeRetrievalMetrics(rootDir: string, behaviors: Agen
  */
 export declare function computeTaskMetrics(taskId: string, area: string, retrieved: string[], canonical: Set<string>): TaskRetrievalMetrics;
 /**
- * Load canonical_docs from all task YAML files.
+ * Load canonical docs from *.task.ts files in tasks/literacy/.
  * Returns a map of taskId → { slugs: Set<string>, area: string }.
  */
 export declare function loadCanonicalDocs(rootDir: string): Map<string, {