@sanity/ailf 2.2.0 → 2.3.1

package/config/rubrics.ts

@@ -201,9 +201,9 @@ export default defineRubrics({
       currency: 0.2,
     },
     "agent-harness": {
-      "agent-output": 0.45,
-      "tool-usage": 0.4,
-      "process-quality": 0.15,
+      "assertion-pass-rate": 0.35,
+      "agent-output": 0.35,
+      "tool-usage": 0.3,
     },
   },
 

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

@@ -126,31 +126,56 @@ export interface FeatureScore {
      * Only present when agentic evaluation data is available.
      */
     actualScore?: number;
+    /**
+     * Assertion pass rate — fraction of structural assertions that passed (0–100).
+     * Only present for modes with javascript assertions (agent-harness, agent-task).
+     */
+    assertionPassRate?: number;
     /**
      * Ceiling score — gold-standard docs injected directly.
      * This is the theoretical maximum score for this area given the current docs.
+     * Set to 0 for modes without with/without-docs variants (agent-harness).
      */
     ceilingScore: number;
     codeCorrectness: number;
+    /**
+     * Generic dimension scores map — all dimensions by kebab-case name (0–100).
+     * Includes the three named fields above plus any mode-specific dimensions
+     * (e.g., agent-output, tool-usage, assertion-pass-rate).
+     * New consumers should read from this map. The named fields are backward-
+     * compatible accessors populated from it.
+     */
+    dimensions?: Record<string, number>;
     docCoverage: number;
     /** Sanity documents used for this feature area's evaluation */
     documents?: DocumentRef[];
     /**
      * Doc Lift — documentation quality contribution (ceiling − floor).
      * Positive when docs help, negative when docs hurt (interference).
+     * Set to 0 for modes without with/without-docs variants (agent-harness).
      */
     docLift: number;
     /**
      * Doc quality gap — room for documentation improvement (100 − ceiling).
      * Lower is better.
+     * Set to 0 for modes without with/without-docs variants (agent-harness).
      */
     docQualityGap: number;
     feature: string;
     /**
      * Floor score — no docs, training data only.
      * The model's inherent knowledge baseline.
+     * Set to 0 for modes without with/without-docs variants (agent-harness).
      */
     floorScore: number;
+    /**
+     * How this score entry was grouped.
+     *   - "feature": by documentation feature area (literacy mode)
+     *   - "task": by individual task ID (agent-harness mode)
+     *   - "aggregate": single aggregate across all tasks
+     * Defaults to "feature" when absent (backward compatibility).
+     */
+    groupType?: "aggregate" | "feature" | "task";
     /**
      * Infrastructure efficiency — actual / ceiling (0.0–1.0).
      * What fraction of documentation quality reaches agents through discovery?

package/dist/adapters/task-sources/content-lake-task-source.js

@@ -28,7 +28,13 @@
  */
 const TASKS_QUERY = /* groq */ `
 *[_type == "ailf.task"
-  && (!defined($areas) || area->areaId.current in $areas)
+  && (
+    !defined($areas)
+    // Current field name
+    || area->areaId.current in $areas
+    // Legacy field name (pre-schema-rename documents)
+    || featureArea->areaId.current in $areas
+  )
   && (!defined($taskIds) || id.current in $taskIds)
   && (
     // Status-based filtering (unified — replaces execution.enabled)
@@ -39,13 +45,15 @@ const TASKS_QUERY = /* groq */ `
     || (defined($taskIds) && status != "archived")
   )
   && (!defined($tags) || count((tags)[@ in $tags]) > 0)
-] | order(area->areaId.current asc, id.current asc) {
+] | order(coalesce(area->areaId.current, featureArea->areaId.current) asc, id.current asc) {
   "taskId": id.current,
-  title,
-  "areaId": area->areaId.current,
-  promptText,
+  // Coalesce current and legacy field names so documents created before
+  // the schema rename are still readable.
+  "title": coalesce(title, description),
+  "areaId": coalesce(area->areaId.current, featureArea->areaId.current),
+  "promptText": coalesce(promptText, taskPrompt),
   docCoverage,
-  "contextDocs": contextDocs[] {
+  "contextDocs": coalesce(contextDocs, canonicalDocs)[] {
     refType,
     "slug": doc->slug.current,
     "docRefId": doc->_id,
@@ -55,7 +63,7 @@ const TASKS_QUERY = /* groq */ `
     perspective,
     reason
   },
-  assertions,
+  "assertions": coalesce(assertions, assert),
   rawAssert,
   baseline,
   tags,

package/dist/commands/calculate-scores.js

@@ -9,18 +9,21 @@ import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
 import { calculateAndWriteScores } from "../pipeline/calculate-scores.js";
+import { addOutputDirOption } from "./shared/options.js";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 export function createCalculateScoresCommand() {
-    return new Command("calculate-scores")
+    const cmd = new Command("calculate-scores")
         .description("Calculate AI Literacy Scores from Promptfoo evaluation results")
         .option("--source <name>", "Documentation source name")
         .argument("[results-path]", "Path to eval-results.json")
         .action(async (resultsPath, opts) => {
         try {
+            const outputDir = resolveOutputDir(opts.outputDir);
             const ctx = createAppContext({
                 rootDir: ROOT,
-                outputDir: resolve(ROOT, "results", "latest"),
+                outputDir,
                 mode: "literacy",
                 noAutoScope: false,
                 skipFetch: true,
@@ -53,4 +56,6 @@ export function createCalculateScoresCommand() {
                 console.error(err.message);
         }
     });
+    addOutputDirOption(cmd);
+    return cmd;
 }

package/dist/commands/capture-list.d.ts

@@ -5,7 +5,7 @@
  * reads each manifest, and prints a summary table sorted by date.
  *
  * Usage:
- *   ailf capture list                          # default: results/captures/
+ *   ailf capture list                          # default: .ailf/results/captures/
  *   ailf capture list ./my-captures            # custom directory
  */
 import { Command } from "commander";

package/dist/commands/capture-list.js

@@ -5,22 +5,25 @@
  * reads each manifest, and prints a summary table sorted by date.
  *
  * Usage:
- *   ailf capture list                          # default: results/captures/
+ *   ailf capture list                          # default: .ailf/results/captures/
  *   ailf capture list ./my-captures            # custom directory
  */
 import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
 import { join, resolve } from "node:path";
 import { Command } from "commander";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 // ---------------------------------------------------------------------------
 // Command factory
 // ---------------------------------------------------------------------------
 export function createCaptureListCommand() {
     return new Command("list")
         .description("List pipeline captures in a directory")
-        .argument("[dir]", "Captures directory (default: results/captures/)")
+        .argument("[dir]", "Captures directory (default: .ailf/results/captures/)")
         .option("-f, --format <fmt>", "Output format: table or json", "table")
         .action(async (dir, opts) => {
-        const captureDir = resolve(dir ?? "results/captures");
+        const captureDir = dir
+            ? resolve(dir)
+            : resolve(resolveOutputDir(), "..", "captures");
         if (!existsSync(captureDir)) {
             console.error(`  No captures directory found at ${captureDir}`);
             console.error("  Run 'ailf pipeline --capture' to create captures.");

package/dist/commands/compare.js

@@ -9,29 +9,31 @@ import { dirname, join, resolve } from "path";
 import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { compare } from "../pipeline/compare.js";
+import { addOutputDirOption } from "./shared/options.js";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 import { DEFAULT_NOISE_THRESHOLD, } from "../pipeline/types.js";
 import { formatComparisonTable } from "../_vendor/ailf-core/index.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 const BASELINES_DIR = join(ROOT, "results", "baselines");
-const SCORE_SUMMARY_PATH = join(ROOT, "results", "latest", "score-summary.json");
 // ---------------------------------------------------------------------------
 // Helpers
 // ---------------------------------------------------------------------------
 export function createCompareCommand() {
-    return new Command("compare")
+    const cmd = new Command("compare")
         .description("Compare two evaluation score summaries")
         .option("-b, --baseline <path>", "Baseline score-summary.json (default: latest baseline)")
-        .option("-e, --experiment <path>", "Experiment score-summary.json (default: results/latest/score-summary.json)")
+        .option("-e, --experiment <path>", "Experiment score-summary.json (default: .ailf/results/latest/score-summary.json)")
         .option("-t, --threshold <n>", "Noise threshold for unchanged classification", parseFloat)
         .option("-o, --output <path>", "Write JSON report to file")
         .option("-f, --format <fmt>", "Output format: table or json", "table")
         .action(async (opts) => {
+        const outputDir = resolveOutputDir(opts.outputDir);
         const threshold = opts.threshold ?? DEFAULT_NOISE_THRESHOLD;
         // Resolve experiment path
         const expPath = opts.experiment
             ? resolve(opts.experiment)
-            : SCORE_SUMMARY_PATH;
+            : join(outputDir, "score-summary.json");
         const experiment = loadSummary(expPath);
         // Resolve baseline path
         let basePath;
@@ -48,7 +50,7 @@ export function createCompareCommand() {
         }
         const baseline = loadSummary(basePath);
         // Try to load grader consistency data for empirical thresholds
-        const consistencyPath = join(ROOT, "results", "latest", "grader-consistency.json");
+        const consistencyPath = join(outputDir, "grader-consistency.json");
         let graderConsistency;
         if (existsSync(consistencyPath) && opts.threshold === undefined) {
             try {
@@ -93,10 +95,12 @@ export function createCompareCommand() {
                 console.log(`  ✅ Comparison report also written to ${opts.output}`);
             }
         }
-        // Write comparison report to results/latest for other steps to consume
-        const latestComparisonPath = join(ROOT, "results", "latest", "comparison-report.json");
+        // Write comparison report to output dir for other steps to consume
+        const latestComparisonPath = join(outputDir, "comparison-report.json");
         writeFileSync(latestComparisonPath, JSON.stringify(report, null, 2));
     });
+    addOutputDirOption(cmd);
+    return cmd;
 }
 function findLatestBaseline() {
     if (!existsSync(BASELINES_DIR))

package/dist/commands/explain-handler.js

@@ -23,6 +23,7 @@ import { TASK_FILE_NAMES } from "../_vendor/ailf-core/index.js";
 import { buildPipelinePlan, buildSimpleCommandPlan, } from "../pipeline/plan.js";
 import { formatPlanConsole, formatPlanJson } from "../pipeline/plan-format.js";
 import { computeResolvedOptions } from "./pipeline-action.js";
+import { getCallerCwd } from "./shared/resolve-output-dir.js";
 import { LiteracyVariant } from "../pipeline/normalize-mode.js";
 // ---------------------------------------------------------------------------
 // Registry
@@ -43,10 +44,10 @@ const EXPLAIN_REGISTRY = {
     "agent-report": {
         description: "Generate an agent behavior observation report from eval results",
         filesCreated: [
-            "results/latest/agent-report.json",
-            "results/latest/agent-report.md",
+            "<outputDir>/agent-report.json",
+            "<outputDir>/agent-report.md",
         ],
-        filesRead: ["results/latest/eval-results.json"],
+        filesRead: ["<outputDir>/eval-results.json"],
         steps: [
             {
                 cacheStatus: "miss",
@@ -82,9 +83,9 @@ const EXPLAIN_REGISTRY = {
     },
     "calculate-scores": {
         description: "Calculate AI Literacy Scores from Promptfoo evaluation results",
-        filesCreated: ["results/latest/score-summary.json"],
+        filesCreated: ["<outputDir>/score-summary.json"],
         filesRead: [
-            "results/latest/eval-results.json",
+            "<outputDir>/eval-results.json",
             "config/rubrics.ts",
             "config/models.ts",
         ],
@@ -104,23 +105,20 @@ const EXPLAIN_REGISTRY = {
             {
                 cacheStatus: "miss",
                 name: "Write summary",
-                reason: "Persist score-summary.json to results/latest/",
+                reason: "Persist score-summary.json to output directory",
                 willRun: true,
             },
         ],
     },
     compare: {
         description: "Compare current evaluation scores against a saved baseline",
-        filesCreated: ["results/latest/comparison-report.json"],
-        filesRead: [
-            "results/latest/score-summary.json",
-            "results/baselines/*.json",
-        ],
+        filesCreated: ["<outputDir>/comparison-report.json"],
+        filesRead: ["<outputDir>/score-summary.json", "results/baselines/*.json"],
         steps: [
             {
                 cacheStatus: "miss",
                 name: "Load current scores",
-                reason: "Read results/latest/score-summary.json",
+                reason: "Read <outputDir>/score-summary.json",
                 willRun: true,
             },
             {
@@ -181,8 +179,8 @@ const EXPLAIN_REGISTRY = {
     },
     "discovery-report": {
         description: "Generate agent discoverability report from agentic retrieval metrics",
-        filesCreated: ["results/latest/discovery-report.md"],
-        filesRead: ["results/latest/score-summary.json"],
+        filesCreated: ["<outputDir>/discovery-report.md"],
+        filesRead: ["<outputDir>/score-summary.json"],
         steps: [
             {
                 cacheStatus: "miss",
@@ -206,7 +204,7 @@ const EXPLAIN_REGISTRY = {
     },
     eval: {
         description: "Run Promptfoo evaluation directly (passthrough — all flags forwarded to promptfoo)",
-        filesCreated: ["results/latest/eval-results.json"],
+        filesCreated: ["<outputDir>/eval-results.json"],
         filesRead: ["promptfooconfig.yaml"],
         steps: [
             {
@@ -280,7 +278,7 @@ const EXPLAIN_REGISTRY = {
     grader: {
         description: "Grader reliability tools (consistency, compare, sensitivity, validate)",
         filesRead: [
-            "results/latest/eval-results.json",
+            "<outputDir>/eval-results.json",
             "config/rubrics.ts",
             "canonical/reference-solutions/",
         ],
@@ -312,7 +310,7 @@ const EXPLAIN_REGISTRY = {
     },
     "measure-retrieval": {
         description: "Measure Sanity text search retrieval quality against canonical document annotations",
-        filesCreated: ["results/latest/retrieval-metrics.json"],
+        filesCreated: ["<outputDir>/retrieval-metrics.json"],
         filesRead: ["tasks/literacy/*.task.ts"],
         steps: [
             {
@@ -337,7 +335,7 @@ const EXPLAIN_REGISTRY = {
     },
     "pr-comment": {
         description: "Generate a markdown PR comment from evaluation scores for CI posting",
-        filesRead: ["results/latest/score-summary.json"],
+        filesRead: ["<outputDir>/score-summary.json"],
         steps: [
             {
                 cacheStatus: "miss",
@@ -355,7 +353,7 @@ const EXPLAIN_REGISTRY = {
     },
     publish: {
         description: "Publish a local evaluation report to the Sanity Content Lake (standalone)",
-        filesRead: ["results/latest/score-summary.json"],
+        filesRead: ["<outputDir>/score-summary.json"],
         steps: [
             {
                 cacheStatus: "miss",
@@ -386,12 +384,12 @@ const EXPLAIN_REGISTRY = {
     "readiness-report": {
         description: "Generate launch readiness checklist for a feature area with threshold evaluation",
         filesRead: [
-            "results/latest/score-summary.json",
-            "results/latest/gap-analysis.json",
+            "<outputDir>/score-summary.json",
+            "<outputDir>/gap-analysis.json",
             "config/thresholds.ts",
             "results/baselines/",
         ],
-        filesCreated: ["results/latest/readiness-report.md"],
+        filesCreated: ["<outputDir>/readiness-report.md"],
         steps: [
             {
                 cacheStatus: "miss",
@@ -603,7 +601,7 @@ function buildInitExplainPlan(actionCommand, rootDir) {
     const configFile = format === "ts"
         ? "ailf.config.ts"
         : `config.${format === "yaml" ? "yaml" : "json"}`;
-    const callerCwd = process.env.AILF_CALLER_CWD ?? process.cwd();
+    const callerCwd = getCallerCwd();
     const targetDir = opts.path ?? ".";
     const ailfDir = `${targetDir}/.ailf`;
     const tasksDir = `${ailfDir}/tasks`;
@@ -664,7 +662,7 @@ function buildBaselineExplainPlan(actionCommand, rootDir) {
         command: `baseline ${subcommand}`,
         description: descriptions[subcommand] ?? `Baseline operation: ${subcommand}`,
         filesCreated: subcommand === "save" ? ["results/baselines/<timestamp>.json"] : [],
-        filesRead: ["results/latest/score-summary.json", "results/baselines/"],
+        filesRead: ["<outputDir>/score-summary.json", "results/baselines/"],
         rootDir,
     });
 }

package/dist/commands/fetch-docs.js

@@ -11,7 +11,8 @@ import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
 import { loadSource } from "../sources.js";
 import { configToSourceOverrides } from "../orchestration/config-to-source-overrides.js";
-import { addSanitySourceOptions } from "./shared/options.js";
+import { addOutputDirOption, addSanitySourceOptions } from "./shared/options.js";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 export function createFetchDocsCommand() {
@@ -31,6 +32,7 @@ export function createFetchDocsCommand() {
         }
     });
     addSanitySourceOptions(cmd);
+    addOutputDirOption(cmd);
     return cmd;
 }
 // ---------------------------------------------------------------------------
@@ -41,7 +43,7 @@ async function executeFetchDocs(opts) {
     // Build a minimal ResolvedConfig for the composition root
     const ctx = createAppContext({
         rootDir: ROOT,
-        outputDir: resolve(ROOT, "results", "latest"),
+        outputDir: resolveOutputDir(opts.outputDir),
         mode: "literacy",
         noAutoScope: false,
         skipFetch: false,

package/dist/commands/generate-configs.js

@@ -9,17 +9,19 @@ import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
 import { GenerateConfigsStep } from "../orchestration/steps/generate-configs-step.js";
+import { addOutputDirOption } from "./shared/options.js";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 export function createGenerateConfigsCommand() {
-    return new Command("generate-configs")
+    const cmd = new Command("generate-configs")
         .description("Generate promptfoo config files from config/models.yaml")
         .option("-s, --source <name>", "Documentation source name")
         .action(async (opts) => {
         try {
             const ctx = createAppContext({
                 rootDir: ROOT,
-                outputDir: resolve(ROOT, "results", "latest"),
+                outputDir: resolveOutputDir(opts.outputDir),
                 mode: "literacy",
                 noAutoScope: false,
                 skipFetch: true,
@@ -58,4 +60,6 @@ export function createGenerateConfigsCommand() {
                 console.error(err.message);
         }
     });
+    addOutputDirOption(cmd);
+    return cmd;
 }

package/dist/commands/pipeline-action.js

@@ -21,6 +21,7 @@ import { buildStepSequence } from "../orchestration/build-step-sequence.js";
 import { orchestratePipeline } from "../orchestration/pipeline-orchestrator.js";
 import { load } from "js-yaml";
 import { parseRepoConfig, } from "../adapters/task-sources/repo-schemas.js";
+import { getCallerCwd, resolveOutputDir } from "./shared/resolve-output-dir.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 // ---------------------------------------------------------------------------
@@ -35,7 +36,7 @@ const VALID_SEARCH_MODES = ["open", "origin-only", "off"];
  */
 export function computeResolvedOptions(opts) {
     // Resolve paths relative to the caller's cwd, not the eval package root
-    const callerCwd = process.env.AILF_CALLER_CWD ?? process.cwd();
+    const callerCwd = getCallerCwd();
     // Validate + normalize mode via the single boundary function.
     // normalizeMode() maps legacy variant names (baseline, agentic, etc.)
     // to canonical mode "literacy" + variant, and throws on invalid input.
@@ -209,23 +210,12 @@ export function computeResolvedOptions(opts) {
     const remote = opts.remote || process.env.AILF_REMOTE === "1";
     const apiUrl = opts.apiUrl ?? process.env.AILF_API_URL ?? "https://ailf-api.sanity.build";
     const apiKey = process.env.AILF_API_KEY ?? undefined;
-    // Output directory: explicit flag → repo-task heuristic → default
+    // Output directory: explicit --output-dir → $CWD/.ailf/results/latest/
     const resolvedRepoTasksPath = opts.repoTasksPath
         ? resolve(callerCwd, opts.repoTasksPath)
         : undefined;
     const resolvedTaskSourceType = resolveTaskSourceType(opts.taskSource);
-    let outputDir;
-    if (opts.outputDir) {
-        outputDir = resolve(callerCwd, opts.outputDir);
-    }
-    else if (resolvedTaskSourceType === "repo" || resolvedRepoTasksPath) {
-        outputDir = resolvedRepoTasksPath
-            ? resolve(resolvedRepoTasksPath, "..", "results", "latest")
-            : resolve(callerCwd, ".ailf", "results", "latest");
-    }
-    else {
-        outputDir = resolve(ROOT, "results", "latest");
-    }
+    const outputDir = resolveOutputDir(opts.outputDir);
     return {
         allowedOriginArgs,
         apiKey,
@@ -310,7 +300,7 @@ export async function executePipeline(cliOpts) {
         }
         const { FileConfigAdapter } = await import("../adapters/config-sources/file-config-adapter.js");
         const { createAppContext } = await import("../composition-root.js");
-        const callerCwd = process.env.AILF_CALLER_CWD ?? process.cwd();
+        const callerCwd = getCallerCwd();
         const adapter = new FileConfigAdapter(cliOpts.config, ROOT);
         const config = await adapter.resolve();
         // Merge CLI-only flags that aren't in the config file.
@@ -323,13 +313,8 @@ export async function executePipeline(cliOpts) {
         if (cliOpts.output) {
             config.outputPath = resolve(callerCwd, cliOpts.output);
         }
-        // Output dir: explicit CLI flag → repo-task heuristic → file-config default
-        if (cliOpts.outputDir) {
-            config.outputDir = resolve(callerCwd, cliOpts.outputDir);
-        }
-        else if (config.repoTasksPath) {
-            config.outputDir = resolve(config.repoTasksPath, "..", "results", "latest");
-        }
+        // Output dir: explicit CLI flag → $CWD/.ailf/results/latest/
+        config.outputDir = resolveOutputDir(cliOpts.outputDir);
         // Create AppContext directly from the merged config so adapters
         // (especially taskSource) are wired from the file config's
         // taskSourceType — not from CLI defaults.
@@ -350,8 +335,7 @@ export async function executePipeline(cliOpts) {
     // cache which never contains .ailf/.
     if (o.remote) {
         const { runRemotePipeline } = await import("./remote-pipeline.js");
-        const callerCwd = process.env.AILF_CALLER_CWD ?? process.cwd();
-        await runRemotePipeline(o, callerCwd);
+        await runRemotePipeline(o, getCallerCwd());
         return;
     }
     // Dry-run: validate only, don't execute steps

package/dist/commands/pipeline.js

@@ -55,7 +55,7 @@ export function createPipelineCommand() {
         .option("--remote", "Submit evaluation to the AILF API instead of running locally", false)
         .option("--api-url <url>", "AILF API base URL (default: https://ailf-api.sanity.build)")
         .option("--capture", "Enable artifact capture for this run", false)
-        .option("--capture-dir <path>", "Base directory for capture output (default: results/captures/)")
+        .option("--capture-dir <path>", "Base directory for capture output (default: .ailf/results/captures/)")
         .option("--no-capture-compress", "Disable tar.gz compression of captures")
         .option("--no-capture-extras", "Exclude mode-specific artifacts from captures")
         .action(async (opts) => {

package/dist/commands/pr-comment.js

@@ -9,10 +9,12 @@ import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
 import { generatePrComment } from "../pipeline/pr-comment.js";
+import { addOutputDirOption } from "./shared/options.js";
+import { resolveOutputDir } from "./shared/resolve-output-dir.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 export function createPrCommentCommand() {
-    return new Command("pr-comment")
+    const cmd = new Command("pr-comment")
         .description("Generate a markdown PR comment from evaluation scores")
         .option("--output <path>", "Write comment to file (default: stdout)")
         .option("--promptfoo-url <url>", "Promptfoo share URL to include")
@@ -20,7 +22,7 @@ export function createPrCommentCommand() {
         try {
             const ctx = createAppContext({
                 rootDir: ROOT,
-                outputDir: resolve(ROOT, "results", "latest"),
+                outputDir: resolveOutputDir(opts.outputDir),
                 mode: "literacy",
                 noAutoScope: false,
                 skipFetch: true,
@@ -48,4 +50,6 @@ export function createPrCommentCommand() {
                 console.error(err.message);
         }
     });
+    addOutputDirOption(cmd);
+    return cmd;
 }

package/dist/commands/publish.d.ts

@@ -21,6 +21,7 @@
 import { Command } from "commander";
 export interface PublishCommandOptions {
     dryRun: boolean;
+    outputDir?: string;
     tag?: string;
 }
 export declare function createPublishCommand(): Command;

package/dist/commands/publish.js

@@ -23,22 +23,27 @@ import { dirname, resolve } from "path";
 import { fileURLToPath } from "url";
 import { Command } from "commander";
 import { createAppContext } from "../composition-root.js";
+import { addOutputDirOption } from "./shared/options.js";
+import { getCallerCwd, resolveOutputDir } from "./shared/resolve-output-dir.js";
 import { buildProvenance, } from "../pipeline/provenance.js";
 import { generateReportTitle } from "../pipeline/report-title.js";
 import { generateReportId, } from "../report-store.js";
 import { withRetry } from "../sinks/retry.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
-const DEFAULT_SUMMARY_PATH = resolve(ROOT, "results", "latest", "score-summary.json");
 export function createPublishCommand() {
-    return new Command("publish")
+    const cmd = new Command("publish")
         .description("Publish a local evaluation report to the Sanity Content Lake")
-        .argument("[summary-path]", "Path to score-summary.json", DEFAULT_SUMMARY_PATH)
+        .argument("[summary-path]", "Path to score-summary.json")
         .option("-t, --tag <tag>", "Label for the published report")
         .option("-n, --dry-run", "Preview the report without writing to Sanity or sinks", false)
         .action(async (summaryPath, opts) => {
-        await runPublishCommand(summaryPath, opts);
+        const outputDir = resolveOutputDir(opts.outputDir);
+        const effectivePath = summaryPath ?? resolve(outputDir, "score-summary.json");
+        await runPublishCommand(effectivePath, outputDir, opts);
     });
+    addOutputDirOption(cmd);
+    return cmd;
 }
 // ---------------------------------------------------------------------------
 // Provenance builder (from score summary, not full pipeline context)
@@ -77,7 +82,7 @@ function buildProvenanceFromSummary(summary) {
 // ---------------------------------------------------------------------------
 // Command implementation
 // ---------------------------------------------------------------------------
-async function runPublishCommand(summaryPath, opts) {
+async function runPublishCommand(summaryPath, outputDir, opts) {
     // Wire up infrastructure via composition root
     const ctx = createAppContext({
         compareEnabled: false,
@@ -87,7 +92,7 @@ async function runPublishCommand(summaryPath, opts) {
         noAutoScope: false,
         noCache: true,
         noRemoteCache: true,
-        outputDir: resolve(ROOT, "results", "latest"),
+        outputDir,
         publishEnabled: true,
         publishTag: opts.tag,
         readinessEnabled: false,
@@ -106,8 +111,7 @@ async function runPublishCommand(summaryPath, opts) {
     // -----------------------------------------------------------------------
     // 1. Resolve and read the score summary
     // -----------------------------------------------------------------------
-    const callerCwd = process.env.AILF_CALLER_CWD ?? process.cwd();
-    const resolvedPath = resolve(callerCwd, summaryPath);
+    const resolvedPath = resolve(getCallerCwd(), summaryPath);
     if (!existsSync(resolvedPath)) {
         console.error(`  ✖ File not found: ${resolvedPath}`);
         console.error();

package/dist/commands/remote-pipeline.js

@@ -88,7 +88,7 @@ export async function runRemotePipeline(opts, rootDir) {
     }
     // 7. Fetch and write output artifacts
     await writeRemoteResults(client, job, {
-        rootDir,
+        outputDir: opts.outputDir,
         outputPath: opts.outputPath,
         apiUrl: opts.apiUrl,
     });