@eigenpal/cli 0.20.44 → 0.20.46

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # @eigenpal/cli
 
+## 0.20.45
+
+### Minor Changes
+
+- 0be0b92: The CLI now follows the chunked Files upload transport, so large on-prem file uploads work without browser-reachable object storage.
+
+## 0.20.45
+
+### Patch Changes
+
+- ded47bf: Clarify how to flag specific dataset review fields and attach field comments with current workflow values. Agent guidance now keeps expected outputs aligned with the full intended workflow output and leaves comparison rules to evaluators.
+
 ## 0.20.42
 
 ### Patch Changes

package/dist/cli.js

@@ -15494,7 +15494,7 @@ var init_human_review = __esm(() => {
   }).refine((value) => value.slice(1).split("/").every((segment) => segment === "*" || !segment.includes("*")), { message: "A wildcard must occupy an entire JSON Pointer segment" });
   HumanReviewCategoricalConfidenceSchema = _enum(["low", "medium", "high"]);
   HumanReviewNumericConfidenceSchema = number2().finite().min(0).max(1);
-  HumanReviewMetadataFromSchema = _enum(["ai.extract"]).describe("Producer adapter. `ai.extract` derives field confidence from `_grounding` and strips that reserved key from review data. Grounding is not stripped unless this is set.");
+  HumanReviewMetadataFromSchema = _enum(["ai.extract"]).describe("Producer adapter. `ai.extract` derives categorical field confidence from `_grounding`, copies valid numeric grounding scores into fieldMetadata.display.score, and strips that reserved key from review data. Grounding is not stripped unless this is set.");
   HumanReviewResolvedConfidenceSchema = union([
     HumanReviewNumericConfidenceSchema,
     HumanReviewCategoricalConfidenceSchema,
@@ -15505,11 +15505,11 @@ var init_human_review = __esm(() => {
     HumanReviewCategoricalConfidenceSchema
   ]);
   HumanReviewFieldMetadataSchema = object({
-    confidence: HumanReviewResolvedConfidenceSchema.optional().describe("Producer-supplied confidence: 0–1 numeric (legacy) or categorical low|medium|high from ai.extract grounding. Not calibrated by Eigenpal."),
+    confidence: HumanReviewResolvedConfidenceSchema.optional().describe("Routing confidence: categorical low|medium|high from ai.extract grounding, or legacy 0–1 numeric when authored. Numeric grounding scores from extract land in display.score instead. Not calibrated by Eigenpal."),
     label: string2().max(500).optional().describe("Short label shown in the review UI"),
     description: string2().max(5000).optional().describe("Longer reviewer guidance for this field"),
     review: _enum(["auto", "always", "never"]).default("auto").optional().describe("Legacy per-field override kept for existing tasks. Prefer selection.fields.review."),
-    display: record(string2(), unknown()).optional().describe("Opaque display metadata preserved for the review UI")
+    display: record(string2(), unknown()).optional().describe("Opaque display metadata preserved for the review UI. When metadataFrom is ai.extract, numeric grounding scores are copied here as display.score; routing confidence stays on confidence as categorical low|medium|high.")
   }).strict();
   HumanReviewFieldPolicySchema = object({
     review: _enum(["always", "never", "skip"]).optional().describe("Always require review, or never/skip review for matching paths"),
@@ -56616,6 +56616,9 @@ var init_eval = __esm(() => {
 });
 
 // ../types/src/git-source/grammar.ts
+function qualifyRunTargetName(name) {
+  return /^wf_[A-Za-z0-9_-]+$/.test(name) ? `workflows.${name}` : name;
+}
 function parseAutomationTarget(target) {
   const parsed = AutomationTargetSchema.parse(target);
   const [rawPackageName, rawRef] = parsed.split("@");
@@ -56634,7 +56637,7 @@ function parseRunTarget(target) {
   const parsed = RunTargetSchema.parse(target);
   if (typeof parsed === "string") {
     const [name, rawRef] = parsed.split("@");
-    const [root, ...segments] = name.split(".");
+    const [root, ...segments] = qualifyRunTargetName(name).split(".");
     const slug = segments.join("/");
     if (root === "agents") {
       return {
@@ -56855,11 +56858,11 @@ var init_grammar = __esm(() => {
       ctx.addIssue({ code: "custom", message: "Run target must be <target> or <target>@<version>" });
       return;
     }
-    const parts = name.split(".");
+    const parts = qualifyRunTargetName(name).split(".");
     if (parts.length < 2) {
       ctx.addIssue({
         code: "custom",
-        message: "Run target must start with agents. or workflows. followed by a slug or id"
+        message: "Run target must be a wf_ id or start with agents. or workflows. followed by a slug or id"
       });
       return;
     }
@@ -207667,10 +207670,29 @@ function resolveRequestUrl(baseUrl, path) {
     return `${base.origin}${requestPath}`;
   return `${base.origin}${basePath}${requestPath}`;
 }
+
+// src/lib/fetch-body.ts
+function isNodeReadableStream(body) {
+  return typeof body.pipe === "function";
+}
+function byteViewToArrayBuffer(view) {
+  if (view.buffer instanceof ArrayBuffer && view.byteOffset === 0 && view.byteLength === view.buffer.byteLength) {
+    return view.buffer;
+  }
+  const copy = new Uint8Array(view.byteLength);
+  copy.set(view);
+  return copy.buffer;
+}
+function toPutBody(body) {
+  if (isNodeReadableStream(body)) {
+    return body;
+  }
+  return byteViewToArrayBuffer(body);
+}
 // package.json
 var package_default = {
   name: "@eigenpal/cli",
-  version: "0.20.44",
+  version: "0.20.46",
   description: "Eigenpal CLI — author and run document workflows from your terminal. Agent-ready.",
   keywords: [
     "eigenpal",
@@ -207863,6 +207885,18 @@ class ApiClient {
     });
     return this.handleResponse(res, url);
   }
+  async putBytes(path, body) {
+    const url = this.buildUrl(path);
+    const stream = isNodeReadableStream(body);
+    const init = {
+      method: "PUT",
+      headers: this.headers({ "Content-Type": "application/octet-stream" }),
+      body,
+      ...stream ? { duplex: "half" } : {}
+    };
+    const res = await fetch(url, init);
+    return this.handleResponse(res, url);
+  }
   async patch(path, body) {
     const url = this.buildUrl(path);
     const res = await fetch(url, {
@@ -211510,7 +211544,7 @@ function registerDatasetReviewRequestCommands(dataset, resolveAutomation) {
       { key: "createdAt", header: "createdAt", format: formatTimestamp }
     ], { ...opts, entityLabel: "review request" });
   }));
-  addJsonFlag(withBaseUrl(reviewRequest.command("create <automation-id>"))).description("Request human review of dataset ground truth. Snapshot examples and poll progress until review is complete.").requiredOption("--title <title>", "Review request title").requiredOption("--example-name <name>", "Example folder name to include (repeatable)", collectRepeatable, []).option("--instructions <text>", "Note shown to the reviewer for the whole request").option("--focus <path>", "Expected-output path to highlight (repeatable)", collectRepeatable, []).option("--focus-reason <spec>", "Reason a focus path needs review, as path=reason (repeatable)", collectRepeatable, []).option("--ignore <path>", "Expected-output path reviewers can skip (repeatable)", collectRepeatable, []).option("--item-note <spec>", "Example-level note seeded at create time, as exampleName=comment (repeatable)", collectRepeatable, []).option("--field-note <spec>", "Field-level note seeded at create time, as exampleName.path=comment (repeatable)", collectRepeatable, []).option("--focus-json <json>", "JSON array of { path, reason? } focus fields").option("--notes-json <json>", "JSON array of { exampleName, comment?, fields?: [{ path, comment }] }").option("--status <draft|open|paused>", "Initial lifecycle status (default: draft)").addHelpText("after", `
+  addJsonFlag(withBaseUrl(reviewRequest.command("create <automation-id>"))).description("Request human review of dataset ground truth. Snapshot examples and poll progress until review is complete.").requiredOption("--title <title>", "Review request title").requiredOption("--example-name <name>", "Example folder name to include (repeatable)", collectRepeatable, []).option("--instructions <text>", "Note shown to the reviewer for the whole request").option("--focus <path>", "Expected-output leaf path to flag across selected examples; parent paths flag every descendant (repeatable)", collectRepeatable, []).option("--focus-reason <spec>", "Reason a focus path needs review, as path=reason (repeatable)", collectRepeatable, []).option("--ignore <path>", "Expected-output path reviewers can skip (repeatable)", collectRepeatable, []).option("--item-note <spec>", "Example-level note seeded at create time, as exampleName=comment (repeatable)", collectRepeatable, []).option("--field-note <spec>", "Field comment only, as exampleName.path=comment; does not flag the field (repeatable)", collectRepeatable, []).option("--focus-json <json>", "JSON array of { path, reason? } focus fields").option("--notes-json <json>", "JSON array of { exampleName, comment?, fields?: [{ path, comment }] }").option("--status <draft|open|paused>", "Initial lifecycle status (default: draft)").addHelpText("after", `
 Examples:
   $ eigenpal workflow dataset review-request create wf_abc123 \\
       --title "Q1 invoice GT review" \\
@@ -211518,12 +211552,21 @@ Examples:
       --instructions "Check IBAN checksums and totals." \\
       --focus vendor.iban --focus-reason 'vendor.iban=OCR often mangles IBANs' \\
       --ignore currency \\
-      --field-note 'invoice-foo.total=off by 0.01 last run' \\
+      --field-note 'invoice-foo.vendor.iban=Workflow currently returns SK12; expected SK34. Verify against the document.' \\
       --status open --json
 
   $ eigenpal workflow dataset review-request list wf_abc123 --status open --json
   $ eigenpal workflow dataset review-request get wf_abc123 dsr_... --json
 
+Keep expected.json one-to-one with the complete intended workflow output,
+including nested objects, arrays, types, and expected file references. Evaluators
+define which values to compare and how; never reshape ground truth for scoring.
+Review snapshots must keep that full dataset shape. Flag only uncertain or
+incorrect leaf paths (for example items.3.amount), not every field or a whole
+array unless every descendant needs review. Focus applies to all selected
+examples. A field note adds context without flagging: include the workflow's
+current value, the expected value, and what the reviewer should verify.
+
 Agent loop: create with focus/ignore and per-example notes → poll \`get --json\`
 until \`.progress.complete\` → record per-field decisions and per-example
 approve/remove with comments → \`pull --out <dir>\` to fetch expected files
@@ -244453,25 +244496,6 @@ import { createReadStream } from "node:fs";
 import { stat as stat3 } from "node:fs/promises";
 import path17 from "node:path";
 
-// src/lib/fetch-body.ts
-function isNodeReadableStream(body) {
-  return typeof body.pipe === "function";
-}
-function byteViewToArrayBuffer(view) {
-  if (view.buffer instanceof ArrayBuffer && view.byteOffset === 0 && view.byteLength === view.buffer.byteLength) {
-    return view.buffer;
-  }
-  const copy = new Uint8Array(view.byteLength);
-  copy.set(view);
-  return copy.buffer;
-}
-function toPutBody(body) {
-  if (isNodeReadableStream(body)) {
-    return body;
-  }
-  return byteViewToArrayBuffer(body);
-}
-
 // src/lib/upload-presigned-multipart.ts
 var MULTIPART_UPLOAD_CONCURRENCY = 4;
 var MULTIPART_PART_MAX_RETRIES = 4;
@@ -244645,6 +244669,79 @@ async function uploadPresignedMultipartParts(options) {
   }
 }
 
+// src/lib/upload-app-chunked.ts
+var HTTP_CHUNKED_MAX_RETRIES = 4;
+var HTTP_CHUNKED_RETRY_BASE_MS = 250;
+function expectedHttpChunkByteLength(totalSize, chunkSizeBytes, chunkIndex, chunkCount) {
+  if (!Number.isInteger(chunkIndex) || chunkIndex < 0 || chunkIndex >= chunkCount) {
+    throw new Error(`chunkIndex must be an integer in [0, ${chunkCount})`);
+  }
+  if (totalSize === 0)
+    return 0;
+  if (chunkIndex < chunkCount - 1)
+    return chunkSizeBytes;
+  const remainder = totalSize % chunkSizeBytes;
+  return remainder === 0 ? chunkSizeBytes : remainder;
+}
+function chunkByteOffset(chunkSizeBytes, chunkIndex) {
+  return chunkIndex * chunkSizeBytes;
+}
+function shouldAbortHttpChunkedUploadSession(options) {
+  return !options.chunksReady;
+}
+function httpChunkedCompleteRetryHint(uploadId) {
+  return `Uploaded chunks remain stored for ${uploadId}; retry complete and do not abort the session.`;
+}
+function annotateHttpChunkedCompleteFailure(uploadId, error) {
+  const hint = httpChunkedCompleteRetryHint(uploadId);
+  if (error instanceof Error) {
+    if (!error.message.includes("retry complete")) {
+      error.message = `${error.message} ${hint}`;
+    }
+    return error;
+  }
+  return new Error(`${String(error)} ${hint}`);
+}
+
+class ChunkUploadHttpError extends Error {
+  status;
+  constructor(status) {
+    super(`Chunk upload failed (${status})`);
+    this.status = status;
+    this.name = "ChunkUploadHttpError";
+  }
+}
+function isRetryableChunkError(error) {
+  if (error instanceof ChunkUploadHttpError)
+    return isTransientHttpStatus(error.status);
+  return true;
+}
+async function uploadHttpChunks(options) {
+  for (let chunkIndex = 0;chunkIndex < options.chunkCount; chunkIndex++) {
+    const length = expectedHttpChunkByteLength(options.totalSize, options.chunkSizeBytes, chunkIndex, options.chunkCount);
+    const start = chunkByteOffset(options.chunkSizeBytes, chunkIndex);
+    let lastError;
+    for (let attempt = 0;attempt <= HTTP_CHUNKED_MAX_RETRIES; attempt++) {
+      try {
+        await options.putChunk({ chunkIndex, start, length });
+        lastError = undefined;
+        break;
+      } catch (error) {
+        lastError = error;
+        if (!isRetryableChunkError(error) || attempt === HTTP_CHUNKED_MAX_RETRIES) {
+          throw error;
+        }
+        await sleep2(HTTP_CHUNKED_RETRY_BASE_MS * 2 ** attempt);
+      }
+    }
+    if (lastError)
+      throw lastError;
+  }
+}
+function sleep2(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+
 // src/lib/upload-reusable-file.ts
 function newIdempotencyKey() {
   return randomUUID();
@@ -244673,6 +244770,9 @@ async function uploadReusableFile(client, input) {
   if (negotiation.transport === "presigned-multipart") {
     return uploadPresignedMultipart(client, source, negotiation);
   }
+  if (negotiation.transport === "http-chunked") {
+    return uploadHttpChunked(client, source, negotiation);
+  }
   let putReady = false;
   try {
     const putBody = source.kind === "buffer" ? toPutBody(new Uint8Array(source.content)) : toPutBody(createReadStream(source.filePath));
@@ -244724,6 +244824,38 @@ async function resolveUploadSource(input) {
     contentType
   };
 }
+async function uploadHttpChunked(client, source, session) {
+  const chunksPath = apiPath(session.chunksUrl);
+  const completePath = apiPath(session.completeUrl);
+  let chunksReady = false;
+  try {
+    await uploadHttpChunks({
+      chunkCount: session.chunkCount,
+      chunkSizeBytes: session.chunkSizeBytes,
+      totalSize: source.size,
+      putChunk: async ({ chunkIndex, start, length }) => {
+        const body = source.kind === "buffer" ? toPutBody(source.content.subarray(start, start + length)) : length === 0 ? toPutBody(new Uint8Array) : toPutBody(createReadStream(source.filePath, { start, end: start + length - 1 }));
+        try {
+          await client.putBytes(`${chunksPath}/${chunkIndex}`, body);
+        } catch (error) {
+          if (error instanceof ApiError)
+            throw new ChunkUploadHttpError(error.status);
+          throw error;
+        }
+      }
+    });
+    chunksReady = true;
+    return await client.post(completePath, {});
+  } catch (error) {
+    if (shouldAbortHttpChunkedUploadSession({ chunksReady })) {
+      await client.delete(apiPath(`/files/uploads/${session.uploadId}`)).catch(() => {
+        return;
+      });
+      throw error;
+    }
+    throw annotateHttpChunkedCompleteFailure(session.uploadId, error);
+  }
+}
 async function uploadPresignedMultipart(client, source, session) {
   const partsPath = apiPath(session.partsUrl);
   const completePath = apiPath(session.completeUrl);