@sanity/ailf 7.3.0 → 7.4.0

package/dist/_vendor/ailf-core/artifact-registry.d.ts

@@ -276,6 +276,24 @@ export declare function diagnosisPathBuilder(): ArtifactObjectPath;
  * `diagnosisVersion` MUST NOT contain `|` (the function rejects that case).
  */
 export declare function encodeDiagnosisPathVersion(diagnosisVersion: string, cardVersion: string): string;
+/**
+ * Convert an entry key (wire format, e.g. `{taskId}::{modelId}`) to a
+ * filename-safe component.
+ *
+ * - `::` → `--` so the wire separator doesn't show up in the filename.
+ * - `/` → `_` so task names like "Content Lake with @sanity/client" don't
+ *   create unintended GCS subdirectories and so `ls` against the per-entry
+ *   directory shows one row per entry.
+ *
+ * Single colons (`:`) are preserved — modelIds like
+ * `anthropic:messages:claude-opus-4-6` are valid GCS object names.
+ *
+ * NOTE: this mapping is not bijective. A taskId containing literal `--`
+ * combined with a modelId could in theory collide with one whose taskId
+ * contains `::`. In practice, production taskIds don't exercise these
+ * combinations.
+ */
+export declare function sanitizeEntryKey(key: string): string;
 /** Test-only reset for the legacy-key warning flag. Not exported publicly. */
 export declare function __resetLegacyTestOutputsWarning(): void;
 /**

package/dist/_vendor/ailf-core/artifact-registry.js

@@ -311,7 +311,7 @@ export function encodeDiagnosisPathVersion(diagnosisVersion, cardVersion) {
  * contains `::`. In practice, production taskIds don't exercise these
  * combinations.
  */
-function sanitizeEntryKey(key) {
+export function sanitizeEntryKey(key) {
     return key.replace(/::/g, "--").replace(/\//g, "_");
 }
 /**

package/dist/_vendor/ailf-core/config-helpers.d.ts

@@ -33,15 +33,37 @@ import type { SinksFile } from "./schemas/sinks.js";
 import type { TestBudgetConfig } from "./schemas/test-budgets.js";
 import type { ModelsConfig } from "./types/index.js";
 import type { GeneralizedTaskDefinition } from "./types/generalized-task.js";
+import type { RepoConfig } from "./types/repo-config.js";
 import type { PackageSurfaceConfig } from "./types/package-surface.js";
 import type { PreflightScoringConfig } from "./types/preflight-scoring.js";
 import type { ModeBase, PresetDefinition } from "./types/plugin-registry.js";
 /**
- * Define an AILF evaluation configuration.
+ * Define a full AILF evaluation configuration (`EvalConfig`).
  *
- * Used in `ailf.config.ts` files for typed configuration authoring.
+ * This is the advanced, standalone config passed via `ailf run --config
+ * <path>`. For the repo-level `.ailf/ailf.config.ts` that `ailf init`
+ * scaffolds and `ailf run` auto-loads, use {@link defineRepoConfig} instead —
+ * its shape is `RepoConfig` (with a `source` object), not `EvalConfig`.
  */
 export declare function defineConfig(config: EvalConfig): EvalConfig;
+/**
+ * Define an AILF repo configuration — the `.ailf/ailf.config.ts` file that
+ * `ailf init` scaffolds and `ailf run` auto-loads.
+ *
+ * Narrows the parameter to `RepoConfig` so authors get full IDE autocomplete
+ * for `source`, `triggers`, `execution`, `owner`, etc. Like the other
+ * `define*` helpers this is a pure identity function — runtime validation
+ * happens later via `RepoConfigSchema` when the pipeline loads the file.
+ *
+ * ```typescript
+ * import { defineRepoConfig } from "@sanity/ailf"
+ *
+ * export default defineRepoConfig({
+ *   source: { projectId: "abc123", dataset: "production" },
+ * })
+ * ```
+ */
+export declare function defineRepoConfig(config: RepoConfig): RepoConfig;
 /**
  * Define an evaluation task with full type narrowing by mode.
  *

package/dist/_vendor/ailf-core/config-helpers.js

@@ -30,13 +30,36 @@ import { CANONICAL_EVAL_MODES } from "../ailf-shared/index.js";
 // Config-level helpers
 // ---------------------------------------------------------------------------
 /**
- * Define an AILF evaluation configuration.
+ * Define a full AILF evaluation configuration (`EvalConfig`).
  *
- * Used in `ailf.config.ts` files for typed configuration authoring.
+ * This is the advanced, standalone config passed via `ailf run --config
+ * <path>`. For the repo-level `.ailf/ailf.config.ts` that `ailf init`
+ * scaffolds and `ailf run` auto-loads, use {@link defineRepoConfig} instead —
+ * its shape is `RepoConfig` (with a `source` object), not `EvalConfig`.
  */
 export function defineConfig(config) {
     return config;
 }
+/**
+ * Define an AILF repo configuration — the `.ailf/ailf.config.ts` file that
+ * `ailf init` scaffolds and `ailf run` auto-loads.
+ *
+ * Narrows the parameter to `RepoConfig` so authors get full IDE autocomplete
+ * for `source`, `triggers`, `execution`, `owner`, etc. Like the other
+ * `define*` helpers this is a pure identity function — runtime validation
+ * happens later via `RepoConfigSchema` when the pipeline loads the file.
+ *
+ * ```typescript
+ * import { defineRepoConfig } from "@sanity/ailf"
+ *
+ * export default defineRepoConfig({
+ *   source: { projectId: "abc123", dataset: "production" },
+ * })
+ * ```
+ */
+export function defineRepoConfig(config) {
+    return config;
+}
 // ---------------------------------------------------------------------------
 // Task-level helpers
 // ---------------------------------------------------------------------------

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

@@ -534,4 +534,4 @@ export declare const EXAMPLES: Record<ExampleType, ExampleRecord>;
 /** GitHub Actions workflow template for AI Literacy evaluation */
 export declare const workflowYaml = "# \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n# AI Literacy Evaluation \u2014 GitHub Actions workflow\n# \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n#\n# Evaluates your documentation quality on every pull request.\n# The AILF CLI reads your .ailf/tasks/ definitions, submits them\n# to the AILF API for evaluation, and writes a score report.\n#\n# Prerequisites:\n#   Add one secret to your repository (Settings \u2192 Secrets \u2192 Actions):\n#     AILF_API_KEY \u2014 your API key (starts with ailf_live_sk_)\n#\n#   @sanity/ailf is published with public npm access, so no npm token is\n#   needed to install the CLI.\n#\n# Customization:\n#   - Narrow the trigger paths to reduce cost (see comment below)\n#   - Check debug_mode for faster iteration (fewer tests)\n#   - See: https://github.com/sanity-labs/ai-literacy-framework\n# \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n\nname: AI Literacy Eval\n\non:\n  pull_request:\n    branches: [main]\n    # Runs on every PR to main by default. To reduce cost:\n    #   paths: [\".ailf/**\", \"docs/**\"]\n\n  workflow_dispatch:\n    inputs:\n      debug_mode:\n        description: \"Run in debug mode (fewer tests, faster iteration)\"\n        type: boolean\n        default: false\n\nconcurrency:\n  group: ailf-eval-${{ github.event.pull_request.number || github.ref }}\n  cancel-in-progress: true\n\njobs:\n  evaluate:\n    name: AI Literacy Evaluation\n    runs-on: ubuntu-latest\n    permissions:\n      contents: read\n      pull-requests: write\n    steps:\n      - uses: actions/checkout@v4\n\n      - name: Run evaluation\n        id: eval\n        env:\n          AILF_API_KEY: ${{ secrets.AILF_API_KEY }}\n          # \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n          # D0037 run provenance envelope \u2014 REPLACE THE OWNER TEAM SLUG\n          # below. Unedited templates produce runs tagged with the literal\n          # placeholder so you can spot them in Studio / BigQuery and fix.\n          #\n          # AILF_CLASSIFICATION values: official | adhoc | experimental |\n          # test | external. External teams should use `adhoc` by default;\n          # `official` is reserved for the core-docs scheduled series.\n          # \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n          AILF_CLASSIFICATION: adhoc\n          AILF_OWNER_TEAM: \"<REPLACE-WITH-YOUR-TEAM-SLUG>\"\n          AILF_OWNER_INDIVIDUAL: ${{ github.actor }}\n        run: |\n          npx @sanity/ailf@latest run --remote \\\n            --output /tmp/ailf-report.md \\\n            ${{ inputs.debug_mode && '--debug' || '' }}\n\n      - name: Post PR comment\n        if: always() && github.event_name == 'pull_request'\n        uses: actions/github-script@v7\n        with:\n          script: |\n            const fs = require('fs');\n\n            // --- Constants ---\n            const MARKER = '<!-- ailf-score-report -->';\n            const HISTORY_START = '<!-- ailf-score-history -->';\n            const HISTORY_END = '<!-- /ailf-score-history -->';\n            const MAX_HISTORY = 3; // keep at most 3 prior runs\n\n            // --- Read new report ---\n            let newReport;\n            try {\n              newReport = fs.readFileSync('/tmp/ailf-report.md', 'utf-8');\n            } catch {\n              newReport = `## \u26A0\uFE0F AI Literacy Evaluation\\n\\nNo report generated. Check the [workflow logs](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}).`;\n            }\n\n            const prNumber = context.issue?.number || context.payload?.pull_request?.number;\n            if (!prNumber) {\n              console.log('No PR number found, skipping comment');\n              return;\n            }\n\n            // --- Find existing comment ---\n            const { data: comments } = await github.rest.issues.listComments({\n              owner: context.repo.owner, repo: context.repo.repo,\n              issue_number: prNumber,\n            });\n            const existing = comments.find(c => c.body?.includes(MARKER));\n\n            // --- Build history from previous comment ---\n            let historyEntries = [];\n            if (existing) {\n              const oldBody = existing.body || '';\n\n              // Collect existing collapsed history entries\n              const histStart = oldBody.indexOf(HISTORY_START);\n              const histEnd = oldBody.indexOf(HISTORY_END);\n              if (histStart !== -1 && histEnd !== -1) {\n                const historyContent = oldBody.slice(histStart + HISTORY_START.length, histEnd).trim();\n                // Split on </details> boundaries to get individual entries\n                if (historyContent) {\n                  historyEntries = historyContent\n                    .split(/<\\/details>\\s*/)\n                    .map(s => s.trim())\n                    .filter(s => s.startsWith('<details>'))\n                    .map(s => s + '\\n</details>');\n                }\n              }\n\n              // Extract the current report (will become the newest history entry)\n              let previousReport = '';\n              if (histStart !== -1) {\n                // Report is between MARKER and the \"Previous runs\" heading (or history section)\n                const markerIdx = oldBody.indexOf(MARKER);\n                // Find the --- separator before history\n                const separatorIdx = oldBody.lastIndexOf('---', histStart);\n                const endIdx = separatorIdx > markerIdx ? separatorIdx : histStart;\n                previousReport = oldBody.slice(markerIdx + MARKER.length, endIdx).trim();\n              } else {\n                // No history yet \u2014 everything after MARKER is the report\n                const markerIdx = oldBody.indexOf(MARKER);\n                if (markerIdx !== -1) {\n                  previousReport = oldBody.slice(markerIdx + MARKER.length).trim();\n                }\n              }\n\n              // Collapse the previous report into a <details> entry\n              if (previousReport) {\n                const scoreMatch = previousReport.match(/Overall:\\s*(\\d+)\\/100/);\n                const score = scoreMatch ? scoreMatch[1] : '?';\n                const dateMatch = previousReport.match(/Generated by.*?\u00B7\\s*([^\u00B7<\\n*]+)/);\n                const date = dateMatch\n                  ? dateMatch[1].trim()\n                  : new Date().toISOString().slice(0, 16).replace('T', ' ') + ' UTC';\n                const entry = `<details>\\n<summary>\uD83D\uDCDC ${date} \u2014 ${score}/100</summary>\\n\\n${previousReport}\\n\\n</details>`;\n                historyEntries.unshift(entry); // newest first\n              }\n\n              // Enforce max history limit\n              historyEntries = historyEntries.slice(0, MAX_HISTORY);\n            }\n\n            // --- Assemble final comment ---\n            const historySection = historyEntries.length > 0\n              ? `\\n\\n---\\n\\n### \uD83D\uDCDC Previous runs\\n\\n${HISTORY_START}\\n${historyEntries.join('\\n\\n')}\\n${HISTORY_END}`\n              : '';\n            const finalBody = `${MARKER}\\n${newReport}${historySection}`;\n\n            if (existing) {\n              await github.rest.issues.updateComment({\n                owner: context.repo.owner, repo: context.repo.repo,\n                comment_id: existing.id, body: finalBody,\n              });\n              console.log(`Updated comment (${historyEntries.length} history entries)`);\n            } else {\n              await github.rest.issues.createComment({\n                owner: context.repo.owner, repo: context.repo.repo,\n                issue_number: prNumber, body: finalBody,\n              });\n              console.log('Created new PR comment');\n            }\n\n      - name: Summary\n        if: always()\n        run: |\n          if [ -f /tmp/ailf-report.md ]; then\n            cat /tmp/ailf-report.md >> \"$GITHUB_STEP_SUMMARY\"\n          else\n            echo \"## \u26A0\uFE0F AI Literacy Evaluation\" >> \"$GITHUB_STEP_SUMMARY\"\n            echo \"\" >> \"$GITHUB_STEP_SUMMARY\"\n            echo \"No report generated. Check the workflow logs.\" >> \"$GITHUB_STEP_SUMMARY\"\n          fi\n";
 /** TypeScript project configuration template (ailf.config.ts) */
-export declare const ailfConfigTs = "/**\n * .ailf/ailf.config.ts \u2014 AI Literacy Framework project configuration.\n *\n * This file configures how the AILF evaluation pipeline runs in this\n * repository. Place it at .ailf/ailf.config.ts in your project root.\n *\n * Evaluations are submitted to the AILF API (ailf-api.sanity.build).\n * The API handles LLM calls, doc fetching, grading, and report\n * publishing. Your repo only needs one secret: AILF_API_KEY.\n *\n * Docs: https://github.com/sanity-labs/ai-literacy-framework\n */\n\nexport default {\n  /**\n   * Documentation source \u2014 which docs are being evaluated.\n   *\n   * This tells the pipeline which Sanity project and dataset contain\n   * the documentation under test. For most users, this is Sanity's own\n   * docs project.\n   */\n  source: {\n    /** Sanity project ID (find yours at sanity.io/manage) */\n    projectId: \"3do82whm\",\n    /** The dataset to query (e.g., \"production\", \"next\") */\n    dataset: \"next\",\n    /**\n     * The public URL of your documentation site.\n     * Used by agentic mode to test agent discoverability.\n     */\n    baseUrl: \"https://www.sanity.io/docs\",\n  },\n\n  /**\n   * Trigger configuration \u2014 when evaluations run automatically.\n   *\n   * Each key is a trigger context. The pipeline checks which trigger\n   * matches the current execution context (PR, merge, schedule, etc.)\n   * and applies its settings.\n   *\n   * Mode options:\n   *   \"validate-only\" \u2014 check that task files parse correctly (fast, no LLM calls)\n   *   \"eval\"          \u2014 run the full evaluation pipeline\n   */\n  triggers: {\n    /** On pull requests: just validate task files parse correctly. */\n    pr: {\n      mode: \"validate-only\",\n    },\n\n    /** When .ailf/ files change in a PR: run a real evaluation. */\n    \"pr-task-change\": {\n      mode: \"eval\",\n      paths: [\".ailf/**\"],\n    },\n\n    /** On merge to main: run evaluation (non-blocking). */\n    main: {\n      mode: \"eval\",\n      blocking: false,\n      notify: true,\n    },\n  },\n}\n";
+export declare const ailfConfigTs = "/**\n * .ailf/ailf.config.ts \u2014 AI Literacy Framework project configuration.\n *\n * This file configures how the AILF evaluation pipeline runs in this\n * repository. Place it at .ailf/ailf.config.ts in your project root.\n *\n * Evaluations are submitted to the AILF API (ailf-api.sanity.build).\n * The API handles LLM calls, doc fetching, grading, and report\n * publishing. Your repo only needs one secret: AILF_API_KEY.\n *\n * Docs: https://github.com/sanity-labs/ai-literacy-framework\n */\n\nimport { defineRepoConfig } from \"@sanity/ailf\"\n\nexport default defineRepoConfig({\n  /**\n   * Documentation source \u2014 which docs are being evaluated.\n   *\n   * This tells the pipeline which Sanity project and dataset contain\n   * the documentation under test. For most users, this is Sanity's own\n   * docs project.\n   */\n  source: {\n    /** Sanity project ID (find yours at sanity.io/manage) */\n    projectId: \"3do82whm\",\n    /** The dataset to query (e.g., \"production\", \"next\") */\n    dataset: \"next\",\n    /**\n     * The public URL of your documentation site.\n     * Used by agentic mode to test agent discoverability.\n     */\n    baseUrl: \"https://www.sanity.io/docs\",\n  },\n\n  /**\n   * Trigger configuration \u2014 when evaluations run automatically.\n   *\n   * Each key is a trigger context. The pipeline checks which trigger\n   * matches the current execution context (PR, merge, schedule, etc.)\n   * and applies its settings.\n   *\n   * Mode options:\n   *   \"validate-only\" \u2014 check that task files parse correctly (fast, no LLM calls)\n   *   \"eval\"          \u2014 run the full evaluation pipeline\n   */\n  triggers: {\n    /** On pull requests: just validate task files parse correctly. */\n    pr: {\n      mode: \"validate-only\",\n    },\n\n    /** When .ailf/ files change in a PR: run a real evaluation. */\n    \"pr-task-change\": {\n      mode: \"eval\",\n      paths: [\".ailf/**\"],\n    },\n\n    /** On merge to main: run evaluation (non-blocking). */\n    main: {\n      mode: \"eval\",\n      blocking: false,\n      notify: true,\n    },\n  },\n})\n";

package/dist/_vendor/ailf-core/examples/index.js

@@ -734,4 +734,4 @@ export const workflowYaml = "# ────────────────
 // TypeScript template exports (for ailf init --output-format ts)
 // ---------------------------------------------------------------------------
 /** TypeScript project configuration template (ailf.config.ts) */
-export const ailfConfigTs = "/**\n * .ailf/ailf.config.ts — AI Literacy Framework project configuration.\n *\n * This file configures how the AILF evaluation pipeline runs in this\n * repository. Place it at .ailf/ailf.config.ts in your project root.\n *\n * Evaluations are submitted to the AILF API (ailf-api.sanity.build).\n * The API handles LLM calls, doc fetching, grading, and report\n * publishing. Your repo only needs one secret: AILF_API_KEY.\n *\n * Docs: https://github.com/sanity-labs/ai-literacy-framework\n */\n\nexport default {\n  /**\n   * Documentation source — which docs are being evaluated.\n   *\n   * This tells the pipeline which Sanity project and dataset contain\n   * the documentation under test. For most users, this is Sanity's own\n   * docs project.\n   */\n  source: {\n    /** Sanity project ID (find yours at sanity.io/manage) */\n    projectId: \"3do82whm\",\n    /** The dataset to query (e.g., \"production\", \"next\") */\n    dataset: \"next\",\n    /**\n     * The public URL of your documentation site.\n     * Used by agentic mode to test agent discoverability.\n     */\n    baseUrl: \"https://www.sanity.io/docs\",\n  },\n\n  /**\n   * Trigger configuration — when evaluations run automatically.\n   *\n   * Each key is a trigger context. The pipeline checks which trigger\n   * matches the current execution context (PR, merge, schedule, etc.)\n   * and applies its settings.\n   *\n   * Mode options:\n   *   \"validate-only\" — check that task files parse correctly (fast, no LLM calls)\n   *   \"eval\"          — run the full evaluation pipeline\n   */\n  triggers: {\n    /** On pull requests: just validate task files parse correctly. */\n    pr: {\n      mode: \"validate-only\",\n    },\n\n    /** When .ailf/ files change in a PR: run a real evaluation. */\n    \"pr-task-change\": {\n      mode: \"eval\",\n      paths: [\".ailf/**\"],\n    },\n\n    /** On merge to main: run evaluation (non-blocking). */\n    main: {\n      mode: \"eval\",\n      blocking: false,\n      notify: true,\n    },\n  },\n}\n";
+export const ailfConfigTs = "/**\n * .ailf/ailf.config.ts — AI Literacy Framework project configuration.\n *\n * This file configures how the AILF evaluation pipeline runs in this\n * repository. Place it at .ailf/ailf.config.ts in your project root.\n *\n * Evaluations are submitted to the AILF API (ailf-api.sanity.build).\n * The API handles LLM calls, doc fetching, grading, and report\n * publishing. Your repo only needs one secret: AILF_API_KEY.\n *\n * Docs: https://github.com/sanity-labs/ai-literacy-framework\n */\n\nimport { defineRepoConfig } from \"@sanity/ailf\"\n\nexport default defineRepoConfig({\n  /**\n   * Documentation source — which docs are being evaluated.\n   *\n   * This tells the pipeline which Sanity project and dataset contain\n   * the documentation under test. For most users, this is Sanity's own\n   * docs project.\n   */\n  source: {\n    /** Sanity project ID (find yours at sanity.io/manage) */\n    projectId: \"3do82whm\",\n    /** The dataset to query (e.g., \"production\", \"next\") */\n    dataset: \"next\",\n    /**\n     * The public URL of your documentation site.\n     * Used by agentic mode to test agent discoverability.\n     */\n    baseUrl: \"https://www.sanity.io/docs\",\n  },\n\n  /**\n   * Trigger configuration — when evaluations run automatically.\n   *\n   * Each key is a trigger context. The pipeline checks which trigger\n   * matches the current execution context (PR, merge, schedule, etc.)\n   * and applies its settings.\n   *\n   * Mode options:\n   *   \"validate-only\" — check that task files parse correctly (fast, no LLM calls)\n   *   \"eval\"          — run the full evaluation pipeline\n   */\n  triggers: {\n    /** On pull requests: just validate task files parse correctly. */\n    pr: {\n      mode: \"validate-only\",\n    },\n\n    /** When .ailf/ files change in a PR: run a real evaluation. */\n    \"pr-task-change\": {\n      mode: \"eval\",\n      paths: [\".ailf/**\"],\n    },\n\n    /** On merge to main: run evaluation (non-blocking). */\n    main: {\n      mode: \"eval\",\n      blocking: false,\n      notify: true,\n    },\n  },\n})\n";

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

@@ -18,7 +18,7 @@ export * from "./examples/index.js";
 export * from "./artifact-registry.js";
 export * from "./batch-signing.js";
 export * from "./constants.js";
-export { defineCanaryTasks, defineConfig, defineFeatures, defineModeBase, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineTestBudgets, defineThresholds, } from "./config-helpers.js";
+export { defineCanaryTasks, defineConfig, defineFeatures, defineModeBase, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRepoConfig, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineTestBudgets, defineThresholds, } from "./config-helpers.js";
 export type { PricingEntry, PromptEntry, SourceEntry, } from "./config-helpers.js";
 export { env } from "./env-helper.js";
 export { NoOpArtifactWriter, NotImplementedError, } from "./ports/artifact-writer.js";

package/dist/_vendor/ailf-core/index.js

@@ -21,7 +21,7 @@ export * from "./constants.js";
 // ---------------------------------------------------------------------------
 // Architecture overhaul — Phase 0 helpers
 // ---------------------------------------------------------------------------
-export { defineCanaryTasks, defineConfig, defineFeatures, defineModeBase, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineTestBudgets, defineThresholds, } from "./config-helpers.js";
+export { defineCanaryTasks, defineConfig, defineFeatures, defineModeBase, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRepoConfig, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineTestBudgets, defineThresholds, } from "./config-helpers.js";
 export { env } from "./env-helper.js";
 export { NoOpArtifactWriter, NotImplementedError, } from "./ports/artifact-writer.js";
 export { assoc, resolveVariantMode, splitTaskVariant, } from "./artifact-capture/association.js";

package/dist/_vendor/ailf-core/ports/context.d.ts

@@ -143,6 +143,12 @@ export interface ResolvedConfig {
     perspectiveOverride?: string;
     /** Sanity studio origin override */
     studioOriginOverride?: string;
+    /**
+     * Documentation base-URL override, sourced from the repo config
+     * `source.baseUrl` (or the `DOC_BASE_URL` env var). Distinct from the
+     * `--url` flag captured in `urls`; `urls[0]` still wins when both are set.
+     */
+    baseUrlOverride?: string;
     /** Sanity document filter args */
     sanityDocumentArgs?: string[];
     /** Report ID that triggered this re-run (flows to provenance.lineage.rerunOf) */

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

@@ -79,6 +79,7 @@ export declare const PipelineRequestSchema: z.ZodObject<{
     }>>;
     source: z.ZodOptional<z.ZodString>;
     sourceReportId: z.ZodOptional<z.ZodString>;
+    studioOrigin: z.ZodOptional<z.ZodString>;
     taskMode: z.ZodOptional<z.ZodEnum<{
         "content-lake": "content-lake";
         inline: "inline";

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

@@ -140,6 +140,7 @@ export const PipelineRequestSchema = z.object({
     searchMode: z.enum(["off", "open", "origin-only"]).optional(),
     source: z.string().optional(),
     sourceReportId: z.string().optional(),
+    studioOrigin: z.string().url().optional(),
     taskMode: z.enum(["content-lake", "inline"]).optional(),
     /**
      * Task-source configuration (W0077 Phase 6h). Mirrors

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

@@ -123,6 +123,8 @@ export interface PipelineRequest {
     searchMode?: "off" | "open" | "origin-only";
     source?: string;
     sourceReportId?: string;
+    /** Studio origin override — maps to `ResolvedConfig.studioOriginOverride`. */
+    studioOrigin?: string;
     taskMode?: "content-lake" | "inline";
     taskSource?: PipelineRequestTaskSource;
     tasks?: string[];

package/dist/adapters/api-client/build-request.d.ts

@@ -54,9 +54,11 @@ export interface RemoteConfigSlice {
     publishEnabled?: boolean;
     publishTag?: string;
     concurrency?: number;
+    baseUrlOverride?: string;
     datasetOverride?: string;
     projectIdOverride?: string;
     perspectiveOverride?: string;
+    studioOriginOverride?: string;
     graderContext?: "rubric-only" | "with-docs";
     graderReplications?: number;
     borderlineReplications?: number;

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

@@ -123,6 +123,15 @@ export async function buildRemoteRequest(options) {
         raw.projectId = config.projectIdOverride;
     if (config.perspectiveOverride)
         raw.perspective = config.perspectiveOverride;
+    if (config.studioOriginOverride) {
+        raw.studioOrigin = config.studioOriginOverride;
+    }
+    // A repo-config `source.baseUrl` rides the existing `urls` channel: the
+    // server maps request.urls -> ResolvedConfig.urls -> source baseUrl
+    // (configToSourceOverrides), mirroring the local path.
+    if (config.baseUrlOverride && !raw.urls) {
+        raw.urls = [config.baseUrlOverride];
+    }
     // Advanced
     if (config.graderContext) {
         raw.graderContext = config.graderContext;

package/dist/adapters/config-sources/cli-config-adapter.d.ts

@@ -2,7 +2,7 @@
  * CliConfigAdapter — resolves pipeline config from Commander CLI flags.
  *
  * This is the default adapter — it wraps the existing option resolution
- * pipeline: PipelineCliOptions → computeResolvedOptions → mapToResolvedConfig.
+ * pipeline: PipelineCliOptions → resolveOptions → mapToResolvedConfig.
  *
  * @see packages/eval/src/commands/pipeline-action.ts — underlying implementation
  */

package/dist/adapters/config-sources/cli-config-adapter.js

@@ -2,11 +2,11 @@
  * CliConfigAdapter — resolves pipeline config from Commander CLI flags.
  *
  * This is the default adapter — it wraps the existing option resolution
- * pipeline: PipelineCliOptions → computeResolvedOptions → mapToResolvedConfig.
+ * pipeline: PipelineCliOptions → resolveOptions → mapToResolvedConfig.
  *
  * @see packages/eval/src/commands/pipeline-action.ts — underlying implementation
  */
-import { computeResolvedOptions } from "../../commands/pipeline-action.js";
+import { resolveOptions } from "../../commands/pipeline-action.js";
 import { mapToResolvedConfig } from "../../orchestration/build-app-context.js";
 export class CliConfigAdapter {
     cliOpts;
@@ -17,7 +17,10 @@ export class CliConfigAdapter {
         this.rootDir = rootDir;
     }
     async resolve() {
-        const resolved = computeResolvedOptions(this.cliOpts);
+        // resolveOptions loads the repo config (`.ailf/ailf.config.ts` etc.)
+        // before mapping CLI flags, so file-sourced source/owner/execution
+        // values reach the resolved config.
+        const resolved = await resolveOptions(this.cliOpts);
         return mapToResolvedConfig(resolved, this.rootDir);
     }
 }

package/dist/commands/explain-handler.js

@@ -22,7 +22,7 @@
 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 { resolveOptions } from "./pipeline-action.js";
 import { getCallerCwd } from "./shared/resolve-output-dir.js";
 import { LiteracyVariant } from "../pipeline/normalize-mode.js";
 // ---------------------------------------------------------------------------
@@ -704,7 +704,7 @@ async function buildPipelineExplainPlan(actionCommand, rootDir) {
         purpose: raw.purpose,
         label: raw.label ?? [],
     };
-    const resolved = computeResolvedOptions(withDefaults);
+    const resolved = await resolveOptions(withDefaults);
     const planOpts = {
         areaOption: resolved.areaOption,
         beforeOption: resolved.beforeOption,

package/dist/commands/init.js

@@ -98,7 +98,7 @@ export async function runInit(opts) {
     const skipped = [];
     // 2. Write project config
     if (format === "ts") {
-        // TypeScript: ailf.config.ts with defineConfig helper
+        // TypeScript: ailf.config.ts wrapped in the defineRepoConfig helper
         const configPath = resolve(ailfDir, "ailf.config.ts");
         if (writeIfNew(configPath, ailfConfigTs, force)) {
             written.push(rel(targetDir, configPath));
@@ -252,6 +252,13 @@ export async function runInit(opts) {
         }
     }
     const taskExt = format === "ts" ? ".task.ts" : format === "yaml" ? ".yaml" : ".json";
+    // Reference the config file we actually wrote (not a hard-coded
+    // config.yaml) so the "Next steps" hints point at a real file.
+    const configFile = format === "ts"
+        ? ".ailf/ailf.config.ts"
+        : format === "yaml"
+            ? ".ailf/config.yaml"
+            : ".ailf/config.json";
     console.log();
     console.log("  Next steps:");
     console.log();
@@ -285,7 +292,7 @@ export async function runInit(opts) {
     console.log("     AILF_API_KEY=... npx @sanity/ailf@latest pipeline --remote --debug");
     console.log();
     console.log("  💡 Or test a remote run against your repo tasks:");
-    console.log("     # First, set `taskSource: { type: repo }` in .ailf/config.yaml");
+    console.log(`     # First, set \`taskSource: { type: repo }\` in ${configFile}`);
     console.log("     AILF_API_KEY=... npx @sanity/ailf@latest run --remote --debug");
     console.log();
     console.log("  💡 Or run locally against your repo tasks:");

package/dist/commands/interpret.js

@@ -18,9 +18,9 @@
 import { dirname, resolve } from "path";
 import { fileURLToPath } from "url";
 import { Command } from "commander";
-import { CARD_REGISTRY_VERSION, diagnosisVersion, } from "../_vendor/ailf-core/index.js";
 import { addOutputDirOption } from "./shared/options.js";
 import { resolveOutputDir } from "./shared/resolve-output-dir.js";
+import { defaultVersionsFromReport } from "./shared/versions-from-report.js";
 // ---------------------------------------------------------------------------
 // Module-level root constant (same pattern as compare.ts)
 // ---------------------------------------------------------------------------
@@ -68,36 +68,6 @@ export function formatCardSummaryLine(card) {
     return `${icon} ${card.cardType}: ${text}`;
 }
 // ---------------------------------------------------------------------------
-// Default versions resolver
-// ---------------------------------------------------------------------------
-/**
- * Derive VersionedInputs from a stored report record.
- *
- * The four-version chain is carried in `report.summary.versions` per the
- * Phase 5 schema, with `diagnosisVersion` sourced from the runner's const.
- * Falls back to hard-coded "unknown" values when the fields are not present
- * (legacy reports without version metadata).
- */
-function defaultVersionsFromReport(report) {
-    const rec = report;
-    const summary = rec.summary;
-    const versions = summary?.versions;
-    return {
-        graderJudgmentsVersion: typeof versions?.graderJudgmentsVersion === "string"
-            ? versions.graderJudgmentsVersion
-            : "unknown",
-        ensembleVersion: typeof versions?.ensembleVersion === "string"
-            ? versions.ensembleVersion
-            : "unknown",
-        diagnosisVersion: typeof versions?.diagnosisVersion === "string"
-            ? versions.diagnosisVersion
-            : diagnosisVersion,
-        cardVersion: typeof versions?.cardVersion === "string"
-            ? versions.cardVersion
-            : CARD_REGISTRY_VERSION,
-    };
-}
-// ---------------------------------------------------------------------------
 // Command factory
 // ---------------------------------------------------------------------------
 /**

package/dist/commands/pipeline-action.d.ts

@@ -13,6 +13,7 @@
 import { type ImpactSummary } from "../pipeline/reverse-mapping.js";
 import type { DebugOptions, EvalMode } from "../pipeline/types.js";
 import { type Diagnosis, type ReportStorePort, type SynthesisCostTelemetry } from "../_vendor/ailf-core/index.d.ts";
+import { type RepoConfig } from "../adapters/task-sources/repo-schemas.js";
 import type { PipelineCliOptions } from "./run.js";
 export interface ResolvedOptions {
     allowedOriginArgs: string[];
@@ -23,6 +24,7 @@ export interface ResolvedOptions {
     compareEnabled: boolean;
     compareThreshold?: number;
     concurrency?: number;
+    baseUrlOverride?: string;
     datasetOverride?: string;
     debug?: DebugOptions;
     dryRun: boolean;
@@ -94,7 +96,7 @@ export interface ResolvedOptions {
  *
  * Exported so the plan builder can call it independently.
  */
-export declare function computeResolvedOptions(opts: PipelineCliOptions): ResolvedOptions;
+export declare function computeResolvedOptions(opts: PipelineCliOptions, repoConfig?: RepoConfig | null): ResolvedOptions;
 /**
  * Determine whether the post-run diagnosis summary hook should fire.
  *
@@ -133,6 +135,13 @@ export declare function runPostPipelineHooks(ctx: {
         run(opts: unknown): Promise<Diagnosis>;
     };
 }): Promise<void>;
+/**
+ * Render a failed `--config` load as a clean CLI diagnostic instead of an
+ * uncaught ZodError stack trace. Mirrors the first-5-issues style of the
+ * Content Lake gates, and appends a cross-schema hint when the file smells
+ * like a `.ailf/ailf.config.ts` (`RepoConfig`) rather than an `EvalConfig`.
+ */
+export declare function formatConfigFileError(err: unknown, filePath: string): string;
 /**
  * Execute the evaluation pipeline.
  *
@@ -142,3 +151,22 @@ export declare function runPostPipelineHooks(ctx: {
  * 4. Delegate to the PipelineOrchestrator
  */
 export declare function executePipeline(cliOpts: PipelineCliOptions): Promise<void>;
+/**
+ * Resolve CLI options into typed ResolvedOptions, loading the repo config
+ * from `<cwd>/.ailf/` first. This is the single async entry point; the pure
+ * `computeResolvedOptions` does the option mapping once the config is loaded.
+ */
+export declare function resolveOptions(opts: PipelineCliOptions): Promise<ResolvedOptions>;
+/**
+ * Load the repo config from `<cwd>/.ailf/`. Probes TS/JS/YAML/JSON in a
+ * fixed precedence order (see `REPO_CONFIG_CANDIDATES`) and returns the
+ * first match, validated against `RepoConfigSchema`. Returns null when no
+ * config file is present, or when the matched file fails to load/parse (a
+ * warning is emitted and the run falls back to defaults + env vars).
+ *
+ * Auto-loads regardless of `--task-source`: the same config file is the
+ * per-environment configuration home for every run (W0077 Phase 6a).
+ * Environment variables still win over file values — that cascade lives in
+ * `computeResolvedOptions`, which receives the parsed config from here.
+ */
+export declare function loadRepoConfig(cwd?: string): Promise<RepoConfig | null>;

package/dist/commands/pipeline-action.js

@@ -20,13 +20,16 @@ import { buildAppContext, parseArtifactUploadEnv, } from "../orchestration/build
 import { buildStepSequence } from "../orchestration/build-step-sequence.js";
 import { orchestratePipeline } from "../orchestration/pipeline-orchestrator.js";
 import { load } from "js-yaml";
+import { ZodError } from "zod";
 import { PLACEHOLDER_OWNER_TEAM, } from "../_vendor/ailf-core/index.js";
 import { parseRepoConfig, } from "../adapters/task-sources/repo-schemas.js";
+import { loadTsConfig } from "../adapters/config-sources/ts-config-loader.js";
 import { getCallerCwd, resolveOutputDir } from "./shared/resolve-output-dir.js";
 // Phase 6 / DIAG-06 — single formatter, single visual contract (D6-04).
 // Import statically so bundlers and type-checkers can verify the export
 // exists at build time rather than deferring to runtime dynamic import.
 import { formatCardSummaryLine } from "./interpret.js";
+import { defaultVersionsFromReport } from "./shared/versions-from-report.js";
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const ROOT = resolve(__dirname, "..", "..");
 // ---------------------------------------------------------------------------
@@ -39,13 +42,14 @@ const VALID_SEARCH_MODES = ["open", "origin-only", "off"];
  *
  * Exported so the plan builder can call it independently.
  */
-export function computeResolvedOptions(opts) {
+export function computeResolvedOptions(opts, repoConfig = null) {
     // Resolve paths relative to the caller's cwd, not the eval package root
     const callerCwd = getCallerCwd();
-    // `.ailf/config.yaml` is the per-environment config home for `ailf run`
-    // (W0077 Phase 6a). Load early so downstream cascades (source, agentic,
-    // owner, output, etc.) can read from it.
-    const repoConfig = loadRepoConfigIfPresent(callerCwd);
+    // The repo config (`.ailf/ailf.config.ts` or `.ailf/config.yaml`, etc.) is
+    // loaded asynchronously by `loadRepoConfig` and injected here so this
+    // function stays pure + synchronous (safe for `--explain`). Downstream
+    // cascades (source, agentic, owner, output, etc.) read from it; env vars
+    // still win at each cascade below.
     // 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.
@@ -198,6 +202,12 @@ export function computeResolvedOptions(opts) {
     const datasetOverride = process.env.SANITY_DATASET ?? repoConfig?.source?.dataset;
     const projectIdOverride = process.env.SANITY_PROJECT_ID ?? repoConfig?.source?.projectId;
     const studioOriginOverride = process.env.SANITY_STUDIO_ORIGIN ?? repoConfig?.source?.studioOrigin;
+    // `source.baseUrl` was parsed by the repo-config schema but never mapped
+    // into the source overrides — it took effect only via the `DOC_BASE_URL`
+    // env var or a named `config/sources.ts` entry (D0022). Map it here with
+    // the same env-wins cascade as the trio above; `configToSourceOverrides`
+    // keeps the explicit `--url` flag (`urls[0]`) ahead of it.
+    const baseUrlOverride = process.env.DOC_BASE_URL ?? repoConfig?.source?.baseUrl;
     // Report store overrides (W0077 Phase 6e — `--report-dataset` and
     // `--report-project` retired). Resolution order:
     //   1. Environment variables (AILF_REPORT_DATASET, AILF_REPORT_PROJECT_ID)
@@ -295,6 +305,7 @@ export function computeResolvedOptions(opts) {
         compareEnabled,
         compareThreshold: opts.threshold,
         concurrency,
+        baseUrlOverride,
         datasetOverride,
         debug,
         dryRun: opts.dryRun,
@@ -512,24 +523,12 @@ export async function runPostPipelineHooks(ctx, result, args) {
             process.stderr.write(`ℹ️ Report not found: ${reportId} — skipping post-summary.\n`);
             return;
         }
-        // Derive version metadata from the stored report (same approach as interpret.ts)
-        const rec = report;
-        const summary = rec.summary;
-        const versions = summary?.versions;
-        const versionedInputs = {
-            graderJudgmentsVersion: typeof versions?.graderJudgmentsVersion === "string"
-                ? versions.graderJudgmentsVersion
-                : "unknown",
-            ensembleVersion: typeof versions?.ensembleVersion === "string"
-                ? versions.ensembleVersion
-                : "unknown",
-            diagnosisVersion: typeof versions?.diagnosisVersion === "string"
-                ? versions.diagnosisVersion
-                : "unknown",
-            cardVersion: typeof versions?.cardVersion === "string"
-                ? versions.cardVersion
-                : "unknown",
-        };
+        // Derive version metadata from the stored report. Shares the single
+        // `defaultVersionsFromReport` helper with `ailf interpret` so the
+        // path-relevant axes fall back to the canonical engine versions instead
+        // of the literal "unknown" that produced `diagnosis-unknown-…` paths
+        // (W0286).
+        const versionedInputs = defaultVersionsFromReport(report);
         // Run the diagnosis
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
         const diagnosis = await runner.run({
@@ -600,6 +599,61 @@ function resolveRepoTasksPath(callerCwd, explicitPath, taskSourceType) {
     return undefined;
 }
 // ---------------------------------------------------------------------------
+// --config file error formatting
+// ---------------------------------------------------------------------------
+/**
+ * Fields that exist on `RepoConfig` (`.ailf/ailf.config.ts`) but NOT on the
+ * `EvalConfig` accepted by `--config`. Their presence — or an object-shaped
+ * `source` where `--config` expects a named-source string — is a strong
+ * signal the user pasted their auto-loaded repo config into a `--config`
+ * file. Everything else (`execution`, `grader`, `publish`, `reportStore`,
+ * `artifacts`, `agentic`, `summary`, `taskSource`, `output`) overlaps between
+ * the two shapes.
+ */
+const REPO_ONLY_CONFIG_KEYS = ["triggers", "owner"];
+/**
+ * Detect whether a `--config` validation error looks like a `RepoConfig`
+ * (`.ailf/ailf.config.ts`) pasted into the wrong place. Two tells:
+ *   - an `unrecognized_keys` issue naming `triggers` or `owner`, or
+ *   - an `invalid_type` issue on `source` (RepoConfig's object `source` vs
+ *     EvalConfig's named-source string).
+ */
+function looksLikeRepoConfig(error) {
+    return error.issues.some((issue) => {
+        if (issue.code === "unrecognized_keys") {
+            const keys = issue.keys ?? [];
+            return keys.some((k) => REPO_ONLY_CONFIG_KEYS.includes(k));
+        }
+        return (issue.code === "invalid_type" &&
+            issue.path.length === 1 &&
+            issue.path[0] === "source");
+    });
+}
+/**
+ * Render a failed `--config` load as a clean CLI diagnostic instead of an
+ * uncaught ZodError stack trace. Mirrors the first-5-issues style of the
+ * Content Lake gates, and appends a cross-schema hint when the file smells
+ * like a `.ailf/ailf.config.ts` (`RepoConfig`) rather than an `EvalConfig`.
+ */
+export function formatConfigFileError(err, filePath) {
+    if (!(err instanceof ZodError)) {
+        const msg = err instanceof Error ? err.message : String(err);
+        return `❌ Failed to load --config file: ${filePath}\n   ${msg}`;
+    }
+    const issues = err.issues
+        .slice(0, 5)
+        .map((i) => `  [${i.path.join(".")}]: ${i.message}`)
+        .join("\n");
+    const more = err.issues.length > 5
+        ? `\n  …and ${err.issues.length - 5} more issue(s)`
+        : "";
+    const lines = [`❌ Invalid --config file: ${filePath}`, `${issues}${more}`];
+    if (looksLikeRepoConfig(err)) {
+        lines.push("", "💡 This looks like a .ailf/ailf.config.ts (RepoConfig), which is a", "   different shape from the EvalConfig that --config expects:", "     • --config (EvalConfig): `source` is the NAME of a source declared", "       in config/sources.ts (a string), plus per-run fields like `areas`,", "       `tasks`, `mode`, and `compare`.", "     • .ailf/ailf.config.ts (RepoConfig): `source` is an object", "       ({ projectId, dataset, baseUrl }) plus repo-only `triggers` and", "       `owner`. It is auto-loaded by every `ailf run` — you don't pass it", "       via --config.", "   If you meant to set repo defaults, place this file at", "   .ailf/ailf.config.ts and drop the --config flag.");
+    }
+    return lines.join("\n");
+}
+// ---------------------------------------------------------------------------
 // Pipeline entry point
 // ---------------------------------------------------------------------------
 /**
@@ -623,7 +677,14 @@ export async function executePipeline(cliOpts) {
         const { createAppContext } = await import("../composition-root.js");
         const callerCwd = getCallerCwd();
         const adapter = new FileConfigAdapter(cliOpts.config, ROOT);
-        const config = await adapter.resolve();
+        let config;
+        try {
+            config = await adapter.resolve();
+        }
+        catch (err) {
+            console.error(formatConfigFileError(err, cliOpts.config));
+            process.exit(1);
+        }
         // When `taskSource.type` is `repo` and no `repoTasksPath` was set in
         // the config file, fall back to `<callerCwd>/.ailf/tasks/` (the
         // location `ailf init` scaffolds). Silent fallback — composition-root
@@ -662,7 +723,7 @@ export async function executePipeline(cliOpts) {
         });
         process.exit(result.success ? 0 : 1);
     }
-    const o = resolveOptions(cliOpts);
+    const o = await resolveOptions(cliOpts);
     console.log(`  📂 Output directory: ${o.outputDir}`);
     // Remote mode — submit to AILF API instead of running locally.
     // Use the caller's working directory (not the package root) because
@@ -724,10 +785,13 @@ function warnIfPlaceholderOwnerTeam() {
         `AILF_OWNER_TEAM) to attribute this run.`);
 }
 /**
- * Resolve CLI options into typed ResolvedOptions.
+ * Resolve CLI options into typed ResolvedOptions, loading the repo config
+ * from `<cwd>/.ailf/` first. This is the single async entry point; the pure
+ * `computeResolvedOptions` does the option mapping once the config is loaded.
  */
-function resolveOptions(opts) {
-    return computeResolvedOptions(opts);
+export async function resolveOptions(opts) {
+    const repoConfig = await loadRepoConfig();
+    return computeResolvedOptions(opts, repoConfig);
 }
 function writePipelineResult(result, outputDir) {
     mkdirSync(outputDir, { recursive: true });
@@ -736,25 +800,69 @@ function writePipelineResult(result, outputDir) {
     console.log(`  📄 Pipeline result: ${resultFile}\n`);
 }
 /**
- * Load `<cwd>/.ailf/config.yaml` if it exists. Returns null when the file
- * is absent or unparseable.
+ * Repo-config filenames probed under `<cwd>/.ailf/`, in resolution
+ * precedence order (highest first). `ailf init` writes `ailf.config.ts`
+ * (default `--format ts`), `config.yaml`, or `config.json` depending on
+ * `--format`; all are honored here.
+ *
+ * TypeScript/JavaScript files load via the same jiti mechanism `.task.ts`
+ * files use (`loadTsConfig`) — there is no second TS-loading path. YAML and
+ * JSON load via `js-yaml` (which also parses JSON). When more than one file
+ * is present the first match wins and the rest are ignored with a warning.
+ */
+const REPO_CONFIG_CANDIDATES = [
+    "ailf.config.ts",
+    "ailf.config.js",
+    "config.ts",
+    "config.js",
+    "config.yaml",
+    "config.yml",
+    "config.json",
+];
+function isTsConfigFile(filename) {
+    return filename.endsWith(".ts") || filename.endsWith(".js");
+}
+/**
+ * Load the repo config from `<cwd>/.ailf/`. Probes TS/JS/YAML/JSON in a
+ * fixed precedence order (see `REPO_CONFIG_CANDIDATES`) and returns the
+ * first match, validated against `RepoConfigSchema`. Returns null when no
+ * config file is present, or when the matched file fails to load/parse (a
+ * warning is emitted and the run falls back to defaults + env vars).
  *
- * Auto-loads regardless of `--task-source`: the same `.ailf/config.yaml` is
- * the per-environment configuration home for every run (W0077 Phase 6a).
- * Subsequent flag-family migrations (6b–6h) read additional fields from
- * this same file via the same loader.
+ * Auto-loads regardless of `--task-source`: the same config file is the
+ * per-environment configuration home for every run (W0077 Phase 6a).
+ * Environment variables still win over file values — that cascade lives in
+ * `computeResolvedOptions`, which receives the parsed config from here.
  */
-function loadRepoConfigIfPresent(cwd) {
-    const configPath = resolve(cwd, ".ailf", "config.yaml");
-    if (!existsSync(configPath))
+export async function loadRepoConfig(cwd = getCallerCwd()) {
+    const ailfDir = resolve(cwd, ".ailf");
+    const present = REPO_CONFIG_CANDIDATES.filter((name) => existsSync(resolve(ailfDir, name)));
+    if (present.length === 0)
         return null;
+    const [chosen, ...shadowed] = present;
+    if (shadowed.length > 0) {
+        console.warn(`  ⚠️  Multiple .ailf config files found; using ${chosen}, ignoring ` +
+            `${shadowed.join(", ")}.`);
+    }
+    const configPath = resolve(ailfDir, chosen);
+    const relPath = `.ailf/${chosen}`;
     try {
-        const raw = readFileSync(configPath, "utf-8");
-        const parsed = load(raw);
-        return parseRepoConfig(parsed);
+        let raw;
+        if (isTsConfigFile(chosen)) {
+            const result = await loadTsConfig(configPath);
+            if (!result.ok) {
+                console.warn(`  ⚠️  Failed to load ${relPath}: ${result.error}`);
+                return null;
+            }
+            raw = result.value;
+        }
+        else {
+            raw = load(readFileSync(configPath, "utf-8"));
+        }
+        return parseRepoConfig(raw, relPath);
     }
     catch (err) {
-        console.warn(`  ⚠️  Failed to parse ${configPath}: ${err instanceof Error ? err.message : String(err)}`);
+        console.warn(`  ⚠️  Failed to parse ${relPath}: ${err instanceof Error ? err.message : String(err)}`);
         return null;
     }
 }

package/dist/commands/remote-pipeline.js

@@ -137,9 +137,11 @@ function toConfigSlice(opts) {
         publishEnabled: opts.publishExplicit ? opts.publishEnabled : undefined,
         publishTag: opts.publishTag,
         concurrency: opts.concurrency,
+        baseUrlOverride: opts.baseUrlOverride,
         datasetOverride: opts.datasetOverride,
         projectIdOverride: opts.projectIdOverride,
         perspectiveOverride: opts.perspectiveOverride,
+        studioOriginOverride: opts.studioOriginOverride,
         graderContext: opts.graderContext,
         graderReplications: opts.graderReplications,
         borderlineReplications: opts.borderlineReplications,

package/dist/commands/shared/versions-from-report.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Derive `VersionedInputs` from a stored report record.
+ *
+ * Shared by both diagnosis emission paths — the `ailf interpret` command and
+ * the post-pipeline summary hook (`runPostPipelineHooks`) — so they build
+ * identical version metadata. Co-locating the logic here is the durable fix
+ * for W0286: the two paths previously inlined separate copies that drifted,
+ * and the hook copy coalesced every axis to the literal `"unknown"`.
+ *
+ * The path-relevant axes fall back to the canonical engine constants, NEVER
+ * `"unknown"`. `diagnosisVersion` is the visible provenance slug in the
+ * diagnosis artifact path (`diagnosis-{diagnosisVersion}-{hash}.json`); a
+ * `"unknown"` slug erases the signal that path is meant to carry. `cardVersion`
+ * feeds the path's content hash and the cache key, so it must also be the real
+ * registry version for cache identity to be correct.
+ *
+ * The two non-path axes (`graderJudgmentsVersion`, `ensembleVersion`) have no
+ * canonical version source today and fall back to `"unknown"`; they affect only
+ * the cache key, not the artifact path. Wiring them to real sources is out of
+ * scope for W0286.
+ */
+import { type VersionedInputs } from "../../_vendor/ailf-core/index.d.ts";
+/**
+ * The four-version chain is carried in `report.summary.versions` per the
+ * Phase 5 schema. When a field is absent (legacy reports, or any report
+ * produced before version metadata was populated), the path-relevant axes
+ * resolve to the canonical constants and the rest to `"unknown"`.
+ */
+export declare function defaultVersionsFromReport(report: unknown): VersionedInputs;

package/dist/commands/shared/versions-from-report.js

@@ -0,0 +1,47 @@
+/**
+ * Derive `VersionedInputs` from a stored report record.
+ *
+ * Shared by both diagnosis emission paths — the `ailf interpret` command and
+ * the post-pipeline summary hook (`runPostPipelineHooks`) — so they build
+ * identical version metadata. Co-locating the logic here is the durable fix
+ * for W0286: the two paths previously inlined separate copies that drifted,
+ * and the hook copy coalesced every axis to the literal `"unknown"`.
+ *
+ * The path-relevant axes fall back to the canonical engine constants, NEVER
+ * `"unknown"`. `diagnosisVersion` is the visible provenance slug in the
+ * diagnosis artifact path (`diagnosis-{diagnosisVersion}-{hash}.json`); a
+ * `"unknown"` slug erases the signal that path is meant to carry. `cardVersion`
+ * feeds the path's content hash and the cache key, so it must also be the real
+ * registry version for cache identity to be correct.
+ *
+ * The two non-path axes (`graderJudgmentsVersion`, `ensembleVersion`) have no
+ * canonical version source today and fall back to `"unknown"`; they affect only
+ * the cache key, not the artifact path. Wiring them to real sources is out of
+ * scope for W0286.
+ */
+import { CARD_REGISTRY_VERSION, diagnosisVersion, } from "../../_vendor/ailf-core/index.js";
+/**
+ * The four-version chain is carried in `report.summary.versions` per the
+ * Phase 5 schema. When a field is absent (legacy reports, or any report
+ * produced before version metadata was populated), the path-relevant axes
+ * resolve to the canonical constants and the rest to `"unknown"`.
+ */
+export function defaultVersionsFromReport(report) {
+    const rec = report;
+    const summary = rec.summary;
+    const versions = summary?.versions;
+    return {
+        graderJudgmentsVersion: typeof versions?.graderJudgmentsVersion === "string"
+            ? versions.graderJudgmentsVersion
+            : "unknown",
+        ensembleVersion: typeof versions?.ensembleVersion === "string"
+            ? versions.ensembleVersion
+            : "unknown",
+        diagnosisVersion: typeof versions?.diagnosisVersion === "string"
+            ? versions.diagnosisVersion
+            : diagnosisVersion,
+        cardVersion: typeof versions?.cardVersion === "string"
+            ? versions.cardVersion
+            : CARD_REGISTRY_VERSION,
+    };
+}

package/dist/index.d.ts

@@ -37,7 +37,7 @@
  * })
  * ```
  */
-export { defineConfig, defineFeatures, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./_vendor/ailf-core/index.d.ts";
+export { defineConfig, defineFeatures, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRepoConfig, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./_vendor/ailf-core/index.d.ts";
 export type { PackageSurfaceConfig, PackageSurfaceEntry, PreflightScoringConfig, PricingEntry, PromptEntry, SourceEntry, } from "./_vendor/ailf-core/index.d.ts";
 export { env } from "./_vendor/ailf-core/index.d.ts";
 export type { AgentHarnessTaskDefinition, CustomTaskDefinition, GeneralizedAssertionDefinition, GeneralizedDocRef, GeneralizedTaskDefinition, GeneralizedTemplatedAssertion, GeneralizedValueAssertion, IdDocRef, KnowledgeProbeTaskDefinition, LiteracyTaskDefinition, MCPServerTaskDefinition, PathDocRef, PerspectiveDocRef, RubricRef, SlugDocRef, TaskCommonFields, TaskDifficulty, TaskOptions, TaskProviderConfig, TaskStatus, } from "./_vendor/ailf-core/index.d.ts";

package/dist/index.js

@@ -40,7 +40,7 @@
 // ---------------------------------------------------------------------------
 // Configuration helpers (define* identity functions for typed authoring)
 // ---------------------------------------------------------------------------
-export { defineConfig, defineFeatures, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./_vendor/ailf-core/index.js";
+export { defineConfig, defineFeatures, defineModels, definePackageSurface, definePreflightScoring, definePricingTable, definePreset, definePrompts, defineRepoConfig, defineRubrics, defineSchedules, defineSinks, defineSources, defineTask, defineThresholds, } from "./_vendor/ailf-core/index.js";
 // ---------------------------------------------------------------------------
 // Environment helper
 // ---------------------------------------------------------------------------

package/dist/orchestration/build-app-context.js

@@ -71,6 +71,7 @@ export function mapToResolvedConfig(opts, rootDir) {
         searchMode: opts.searchMode ?? "open",
         concurrency: opts.concurrency,
         promptfooUrl: opts.promptfooUrl,
+        baseUrlOverride: opts.baseUrlOverride,
         datasetOverride: opts.datasetOverride,
         projectIdOverride: opts.projectIdOverride,
         perspectiveOverride: opts.perspectiveOverride,

package/dist/orchestration/config-to-source-overrides.js

@@ -5,11 +5,15 @@
  * with typed overrides instead of relying on process.env.
  */
 export function configToSourceOverrides(config) {
+    // The explicit `--url` flag (captured in `urls`) wins over a repo-config
+    // `source.baseUrl` (captured in `baseUrlOverride`); both feed the same
+    // `SourceOverrides.baseUrl` the doc fetcher reads.
+    const baseUrl = config.urls?.[0] ?? config.baseUrlOverride;
     return {
         ...(config.allowedOrigins?.length
             ? { allowedOrigins: config.allowedOrigins }
             : {}),
-        ...(config.urls?.[0] ? { baseUrl: config.urls[0] } : {}),
+        ...(baseUrl ? { baseUrl } : {}),
         ...(config.datasetOverride ? { dataset: config.datasetOverride } : {}),
         ...(config.sanityDocumentArgs?.length
             ? { documentIds: config.sanityDocumentArgs }

package/dist/pipeline/calculate-scores.js

@@ -1479,9 +1479,15 @@ export async function calculateAndWriteScores(options) {
                 logger: log,
             });
             // Mutate-in-place so subsequent steps (validateGraderJudgmentsCalibration,
-            // persist) see the consensus-merged scores.
+            // persist) see the consensus-merged scores. Snapshot first: the runner's
+            // no-borderline fast path returns the SAME array reference it received,
+            // so `regraded` may alias `judgments`. Truncating `judgments` would then
+            // empty `regraded` before the spread reads it, silently wiping every
+            // judgment (extract N, persist 0) — the divergence the post-persist guard
+            // aborts on. Copying breaks the alias regardless of what the runner returns.
+            const merged = [...regraded];
             judgments.length = 0;
-            judgments.push(...regraded);
+            judgments.push(...merged);
             if (consistencyByJudgment.size > 0) {
                 log.info(`Borderline consensus merged ${consistencyByJudgment.size} judgment(s)`);
             }

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

@@ -68,7 +68,7 @@ export function mapRequestToConfig(request, rootDir) {
         taskSourceType: mapTaskSourceType(request.taskSource?.type, request.taskMode),
         outputPath: undefined,
         promptfooUrl: undefined,
-        studioOriginOverride: undefined,
+        studioOriginOverride: request.studioOrigin,
         sanityDocumentArgs: undefined,
         sourceReportId: request.sourceReportId,
         beforeOption: undefined,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sanity/ailf",
-  "version": "7.3.0",
+  "version": "7.4.0",
   "private": false,
   "publishConfig": {
     "access": "public"