bitfab-cli 0.2.439 → 0.2.440

package/dist/index.js

@@ -4974,7 +4974,7 @@ var require_compile = __commonJS({
       const schOrFunc = root.refs[ref];
       if (schOrFunc)
         return schOrFunc;
-      let _sch = resolve5.call(this, root, ref);
+      let _sch = resolve6.call(this, root, ref);
       if (_sch === void 0) {
         const schema = (_a3 = root.localRefs) === null || _a3 === void 0 ? void 0 : _a3[ref];
         const { schemaId } = this.opts;
@@ -5001,7 +5001,7 @@ var require_compile = __commonJS({
     function sameSchemaEnv(s1, s2) {
       return s1.schema === s2.schema && s1.root === s2.root && s1.baseId === s2.baseId;
     }
-    function resolve5(root, ref) {
+    function resolve6(root, ref) {
       let sch;
       while (typeof (sch = this.refs[ref]) == "string")
         ref = sch;
@@ -5632,7 +5632,7 @@ var require_fast_uri = __commonJS({
       }
       return uri;
     }
-    function resolve5(baseURI, relativeURI, options) {
+    function resolve6(baseURI, relativeURI, options) {
       const schemelessOptions = options ? Object.assign({ scheme: "null" }, options) : { scheme: "null" };
       const { parsed: baseParsed, malformedAuthorityOrPort: baseMalformed } = parseWithStatus(baseURI, schemelessOptions);
       const { parsed: relativeParsed, malformedAuthorityOrPort: relativeMalformed } = parseWithStatus(relativeURI, schemelessOptions);
@@ -5916,7 +5916,7 @@ var require_fast_uri = __commonJS({
     var fastUri = {
       SCHEMES,
       normalize,
-      resolve: resolve5,
+      resolve: resolve6,
       resolveComponent,
       equal,
       serialize,
@@ -8914,6 +8914,10 @@ import { cancel as cancel3 } from "@clack/prompts";
 import { execSync as execSync2 } from "child_process";
 import * as p7 from "@clack/prompts";
 
+// ../bitfab-plugin-lib/dist/bitfabToolHandlers.js
+import { readFile } from "fs/promises";
+import { resolve } from "path";
+
 // ../node_modules/.pnpm/zod@4.6.5/node_modules/zod/v4/classic/external.js
 var external_exports = {};
 __export(external_exports, {
@@ -13940,7 +13944,7 @@ var recursive = /* @__PURE__ */ new WeakMap();
 var NONE = 0;
 var ASSUMED = 1;
 var PROVEN = 2;
-function isRecursive(inst, stack, resolve5) {
+function isRecursive(inst, stack, resolve6) {
   const cached2 = recursive.get(inst);
   if (cached2 !== void 0)
     return cached2 ? PROVEN : NONE;
@@ -13950,7 +13954,7 @@ function isRecursive(inst, stack, resolve5) {
   let result = NONE;
   const check2 = (child) => {
     if (result !== PROVEN && child?._zod) {
-      const answer = isRecursive(child, stack, resolve5);
+      const answer = isRecursive(child, stack, resolve6);
       if (answer > result)
         result = answer;
     }
@@ -13961,7 +13965,7 @@ function isRecursive(inst, stack, resolve5) {
       const desc = Object.getOwnPropertyDescriptor(sh, key);
       if (spread && !desc.enumerable)
         continue;
-      const child = desc.get ? ASSUMED : desc.value?._zod ? isRecursive(desc.value, stack, resolve5) : NONE;
+      const child = desc.get ? ASSUMED : desc.value?._zod ? isRecursive(desc.value, stack, resolve6) : NONE;
       if (child > answer)
         answer = child;
     }
@@ -14025,7 +14029,7 @@ function isRecursive(inst, stack, resolve5) {
       break;
     // `$ZodLazy` caches its inner on the def, so a resolved edge is followed exactly
     case "lazy": {
-      const inner = def._cachedInner ?? (resolve5 ? inst._zod.innerType : void 0);
+      const inner = def._cachedInner ?? (resolve6 ? inst._zod.innerType : void 0);
       merge2(inner ? isRecursive(inner, stack, false) : ASSUMED);
       break;
     }
@@ -29253,17 +29257,29 @@ var getScenario = {
 var saveScenario = {
   name: "save_scenario",
   title: "Save Scenario",
-  description: "Update one scenario, found by its id or by a trace it was built from, for example to add history the run had available, record data it was missing, or correct the trigger. name, description, trigger, and data each replace the saved value whole, and fields left out are kept. To add to the data without resending it, use appendData instead of data: its history entries go after the saved ones and its missing items after the saved missing items, in one step that cannot lose another caller's append made at the same time. Send data or appendData, not both. trigger, data, and appendData entries must keep the shape get_scenario returns. A history entry's source names the span its text came from, on any trace linked to the scenario (a trace that finds it by traceId is one); set it to null or leave it out when the text is not in such a trace, and the entry still counts as data the run had but cannot be cited as evidence when the scenario is turned into assertions. A source naming a span on any other trace is refused. The trigger and data summaries are rebuilt from what is saved. It only updates a scenario that already exists; a trace with no scenario, or a scenario id that does not exist, is refused. Returns the saved scenario.",
+  description: "Create or update one scenario. To create a scenario, or to replace its records, send traceId with records or recordsPath: the records your own system holds that relate to that trace's run, as of the run. Bitfab builds the trigger and data from the trace when the trace has no scenario yet (so far only for traces of Noah's noah-agent-turn function; any other trace with no scenario is refused), then looks for each record's quote, and failing that its text, in the inputs and outputs of the trace's spans. A record found in a span cites that span and can back an assertion; a record found in no span is kept as context the run never saw. A trace that is a replay is refused, because a scenario belongs on the original. records and recordsPath replace the saved records whole, need traceId rather than scenarioId, and cannot be sent with any other field. To edit a scenario that already exists, find it by its id or by a trace it was built from and send name, description, trigger, data, or appendData, for example to add history the run had available, record data it was missing, or correct the trigger. name, description, trigger, and data each replace the saved value whole, and fields left out are kept. To add to the data without resending it, use appendData instead of data: its history entries go after the saved ones and its missing items after the saved missing items, in one step that cannot lose another caller's append made at the same time. Send data or appendData, not both. trigger, data, and appendData entries must keep the shape get_scenario returns. A history entry's source names the span its text came from, on any trace linked to the scenario (a trace that finds it by traceId is one); set it to null or leave it out when the text is not in such a trace, and the entry still counts as data the run had but cannot be cited as evidence when the scenario is turned into assertions. A source naming a span on any other trace is refused. The trigger and data summaries are rebuilt from what is saved. An edit to a trace with no scenario, or to a scenario id that does not exist, is refused. Returns the saved scenario.",
   inputSchema: {
     ...scenarioTargetSchema,
     name: external_exports.string().min(1).optional().describe("The scenario's name."),
     description: external_exports.string().min(1).optional().describe("The scenario's description."),
     trigger: external_exports.preprocess(parseJsonString, external_exports.record(external_exports.string(), external_exports.unknown())).optional().describe("The whole trigger object, in the shape get_scenario returns: what started the run."),
-    data: external_exports.preprocess(parseJsonString, external_exports.record(external_exports.string(), external_exports.unknown())).optional().describe("The whole data object, in the shape get_scenario returns: topic, history (earlier facts in time order, each with its source span), and missing (what the run did not have). Replaces the saved data. Leave it out when sending appendData."),
+    data: external_exports.preprocess(parseJsonString, external_exports.record(external_exports.string(), external_exports.unknown())).optional().describe("The whole data object, in the shape get_scenario returns: topic, history (earlier facts in time order, each with its source span), and missing (what the run did not have). Replaces the saved data, except that the saved records are kept when it has no records field. Leave it out when sending appendData."),
     appendData: external_exports.preprocess(parseJsonString, external_exports.object({
       history: external_exports.array(external_exports.record(external_exports.string(), external_exports.unknown())).optional(),
       missing: external_exports.array(external_exports.string()).optional()
-    })).optional().describe("Data to add to the saved data: history entries (same shape as data.history, source optional, in time order, after the saved ones) and missing items. At least one of the two must be non-empty. Leave data out when sending this.")
+    })).optional().describe("Data to add to the saved data: history entries (same shape as data.history, source optional, in time order, after the saved ones) and missing items. At least one of the two must be non-empty. Leave data out when sending this."),
+    records: external_exports.preprocess(parseJsonString, external_exports.array(external_exports.record(external_exports.string(), external_exports.unknown()))).optional().describe(`The records related to this run, each { kind, ref, at, asOf, title, text, quote }. kind is one of "message", "contact", "preference", "memory", "email", "calendar_event", "action_item". ref is your own pointer to the row, such as "Input:234605". at is when the record came to exist, or null. asOf is "run" when the record is as it was at the run, or "now" when no history is kept for it. quote is a short exact string to look for in the run's spans, or null to look for the whole text. Prefer recordsPath, so the records never pass through the conversation. Send records or recordsPath, not both.`),
+    recordsPath: external_exports.string().min(1).optional().describe("Path to a local JSON file holding the records, either as an array or as an object with a records array. The Bitfab plugin reads the file and sends its records, so they never pass through the conversation. Send records or recordsPath, not both.")
+  }
+};
+var generateScenarioAssertions = {
+  name: "generate_scenario_assertions",
+  title: "Generate Scenario Assertions",
+  description: "Turn a person's label on the scenario built from one trace into up to four assertions, saved as pending agent assertions on that trace for review. The note says what the run got right or wrong; the verdict says whether the run was right overall. Each assertion cites the spans that settle its values, from the scenario's trigger, history, and records the run saw; records the run never saw are context only. A note that gives nothing a reviewer could check against a replay yields no assertions. A trace that is a replay, or a trace with no scenario, is refused. Returns the saved assertions.",
+  inputSchema: {
+    traceId: external_exports.uuid().describe("ID of the trace whose run the person labeled. Its scenario supplies the context, and the assertions are saved on this trace."),
+    verdict: external_exports.preprocess(parseJsonString, external_exports.boolean()).optional().describe("true when the person said the run was right, false when wrong. Leave it out when they gave no verdict."),
+    note: external_exports.string().min(1).describe("What the person said was right or wrong about the run, in their words.")
   }
 };
 var ALL_TOOL_CONTRACTS = [
@@ -29320,7 +29336,8 @@ var ALL_TOOL_CONTRACTS = [
   getSimPlan,
   saveSimPlan,
   getScenario,
-  saveScenario
+  saveScenario,
+  generateScenarioAssertions
 ];
 var TOOL_NAMES = ALL_TOOL_CONTRACTS.map((contract) => contract.name).sort();
 
@@ -29718,6 +29735,36 @@ function withLegacyExperimentKeys(args) {
     ...args.experimentIds === void 0 ? {} : { testRunIds: args.experimentIds }
   };
 }
+function recordsInFile(parsed) {
+  if (Array.isArray(parsed)) {
+    return parsed;
+  }
+  if (typeof parsed === "object" && parsed !== null && "records" in parsed && Array.isArray(parsed.records)) {
+    return parsed.records;
+  }
+  return null;
+}
+async function readRecordsFile(recordsPath) {
+  const fullPath = resolve(recordsPath);
+  let contents;
+  try {
+    contents = await readFile(fullPath, "utf8");
+  } catch (error62) {
+    return {
+      error: `could not read recordsPath ${fullPath}: ${error62 instanceof Error ? error62.message : String(error62)}`
+    };
+  }
+  let parsed;
+  try {
+    parsed = JSON.parse(contents);
+  } catch {
+    return { error: `recordsPath ${fullPath} is not JSON.` };
+  }
+  const records = recordsInFile(parsed);
+  return records === null ? {
+    error: `recordsPath ${fullPath} holds neither an array of records nor an object with a records array.`
+  } : { records };
+}
 function createBitfabToolHandlers(platform2, getConfig2, pluginVersion) {
   const proxy = new McpProxy(platform2, pluginVersion);
   function proxyToolCall(toolName, args) {
@@ -29760,10 +29807,25 @@ function createBitfabToolHandlers(platform2, getConfig2, pluginVersion) {
       notes
     });
   }
+  async function handleSaveScenario(args) {
+    const { recordsPath, ...rest } = args;
+    if (recordsPath === void 0) {
+      return proxyToolCall("save_scenario", rest);
+    }
+    if (rest.records !== void 0) {
+      return errorResult("Error saving scenario: send records or recordsPath, not both.");
+    }
+    const read = await readRecordsFile(recordsPath);
+    if ("error" in read) {
+      return errorResult(`Error saving scenario: ${read.error}`);
+    }
+    return proxyToolCall("save_scenario", { ...rest, records: read.records });
+  }
   const specialHandlers = {
     [saveAgentLabels.name]: handleSaveAgentLabels,
     [saveExperiment.name]: handleSaveExperiment,
-    [saveExperimentGroup.name]: handleSaveExperimentGroup
+    [saveExperimentGroup.name]: handleSaveExperimentGroup,
+    [saveScenario.name]: handleSaveScenario
   };
   const handlers = ALL_TOOL_CONTRACTS.map((contract) => ({
     contract,
@@ -29952,7 +30014,7 @@ function exists(p9) {
     return false;
   }
 }
-function readFile(p9) {
+function readFile2(p9) {
   try {
     return fs6.readFileSync(p9, "utf-8");
   } catch {
@@ -30003,7 +30065,7 @@ function expandGlob(rootDir, pattern) {
   });
 }
 function readPnpmWorkspacePackages(rootDir) {
-  const content = readFile(path5.join(rootDir, "pnpm-workspace.yaml"));
+  const content = readFile2(path5.join(rootDir, "pnpm-workspace.yaml"));
   if (!content) {
     return [];
   }
@@ -30021,7 +30083,7 @@ function readPnpmWorkspacePackages(rootDir) {
   return patterns.flatMap((p9) => expandGlob(rootDir, p9));
 }
 function readPackageJsonWorkspaces(rootDir) {
-  const content = readFile(path5.join(rootDir, "package.json"));
+  const content = readFile2(path5.join(rootDir, "package.json"));
   if (!content) {
     return [];
   }
@@ -30039,7 +30101,7 @@ function readPackageJsonWorkspaces(rootDir) {
   }
 }
 function readUvWorkspaceMembers(rootDir) {
-  const content = readFile(path5.join(rootDir, "pyproject.toml"));
+  const content = readFile2(path5.join(rootDir, "pyproject.toml"));
   if (!content) {
     return [];
   }
@@ -30058,7 +30120,7 @@ function readUvWorkspaceMembers(rootDir) {
   return patterns.flatMap((p9) => expandGlob(rootDir, p9));
 }
 function readGoWorkModules(rootDir) {
-  const content = readFile(path5.join(rootDir, "go.work"));
+  const content = readFile2(path5.join(rootDir, "go.work"));
   if (!content) {
     return [];
   }
@@ -30713,7 +30775,7 @@ async function closeLoginWindow(signInUrl) {
     const appName = path15.match(/\/([^/]+)\.app\//)?.[1];
     return appName === void 0 ? [] : [appName];
   }));
-  await Promise.all(appNames.map((appName) => new Promise((resolve5) => {
+  await Promise.all(appNames.map((appName) => new Promise((resolve6) => {
     const script = [
       "on run argv",
       "  set loginPrefix to item 1 of argv",
@@ -30736,9 +30798,9 @@ async function closeLoginWindow(signInUrl) {
       "end run"
     ].join("\n");
     try {
-      execFile("osascript", ["-e", script, `${url2.origin}${url2.pathname}?`, marker], { timeout: 4e3 }, () => resolve5());
+      execFile("osascript", ["-e", script, `${url2.origin}${url2.pathname}?`, marker], { timeout: 4e3 }, () => resolve6());
     } catch {
-      resolve5();
+      resolve6();
     }
   })));
 }
@@ -31037,7 +31099,7 @@ import { join } from "path";
 // ../bitfab-plugin-lib/dist/commands/replayProgress.js
 import { spawn as spawn2 } from "child_process";
 import { createWriteStream, mkdirSync as mkdirSync4, rmSync as rmSync2 } from "fs";
-import { dirname as dirname3, resolve as resolve3 } from "path";
+import { dirname as dirname3, resolve as resolve4 } from "path";
 import { createInterface } from "readline";
 
 // ../bitfab-plugin-lib/dist/replayCloudResult.js
@@ -31061,7 +31123,7 @@ import { dirname } from "path";
 // ../bitfab-plugin-lib/dist/replayRunDir.js
 import { randomUUID } from "crypto";
 import { appendFileSync, existsSync, mkdirSync as mkdirSync2, readFileSync as readFileSync2, statSync } from "fs";
-import { dirname as dirname2, join as join2, resolve } from "path";
+import { dirname as dirname2, join as join2, resolve as resolve2 } from "path";
 var EXPERIMENT_ID_KEYS = [
   "experimentId",
   "experiment_id",
@@ -31087,7 +31149,7 @@ function extractExperimentId(result) {
 
 // ../bitfab-plugin-lib/dist/replayResultResolver.js
 import { readFileSync as readFileSync3, rmSync } from "fs";
-import { resolve as resolve2 } from "path";
+import { resolve as resolve3 } from "path";
 var replayResultSchema = external_exports.object({
   items: external_exports.array(external_exports.unknown()),
   experimentId: external_exports.string().optional(),
@@ -46276,7 +46338,7 @@ var Protocol = class {
           return;
         }
         const pollInterval = task2.pollInterval ?? this._options?.defaultTaskPollInterval ?? 1e3;
-        await new Promise((resolve5) => setTimeout(resolve5, pollInterval));
+        await new Promise((resolve6) => setTimeout(resolve6, pollInterval));
         options?.signal?.throwIfAborted();
       }
     } catch (error62) {
@@ -46293,7 +46355,7 @@ var Protocol = class {
    */
   request(request, resultSchema, options) {
     const { relatedRequestId, resumptionToken, onresumptiontoken, task, relatedTask } = options ?? {};
-    return new Promise((resolve5, reject) => {
+    return new Promise((resolve6, reject) => {
       const earlyReject = (error62) => {
         reject(error62);
       };
@@ -46371,7 +46433,7 @@ var Protocol = class {
           if (!parseResult.success) {
             reject(parseResult.error);
           } else {
-            resolve5(parseResult.data);
+            resolve6(parseResult.data);
           }
         } catch (error62) {
           reject(error62);
@@ -46632,12 +46694,12 @@ var Protocol = class {
       }
     } catch {
     }
-    return new Promise((resolve5, reject) => {
+    return new Promise((resolve6, reject) => {
       if (signal.aborted) {
         reject(new McpError(ErrorCode.InvalidRequest, "Request cancelled"));
         return;
       }
-      const timeoutId = setTimeout(resolve5, interval);
+      const timeoutId = setTimeout(resolve6, interval);
       signal.addEventListener("abort", () => {
         clearTimeout(timeoutId);
         reject(new McpError(ErrorCode.InvalidRequest, "Request cancelled"));
@@ -47728,7 +47790,7 @@ var McpServer = class {
     let task = createTaskResult.task;
     const pollInterval = task.pollInterval ?? 5e3;
     while (task.status !== "completed" && task.status !== "failed" && task.status !== "cancelled") {
-      await new Promise((resolve5) => setTimeout(resolve5, pollInterval));
+      await new Promise((resolve6) => setTimeout(resolve6, pollInterval));
       const updatedTask = await extra.taskStore.getTask(taskId);
       if (!updatedTask) {
         throw new McpError(ErrorCode.InternalError, `Task ${taskId} not found during polling`);
@@ -48392,12 +48454,12 @@ var StdioServerTransport = class {
     this.onclose?.();
   }
   send(message) {
-    return new Promise((resolve5) => {
+    return new Promise((resolve6) => {
       const json2 = serializeMessage(message);
       if (this._stdout.write(json2)) {
-        resolve5();
+        resolve6();
       } else {
-        this._stdout.once("drain", resolve5);
+        this._stdout.once("drain", resolve6);
       }
     });
   }
@@ -49031,7 +49093,7 @@ async function runClaudeAnalyzeRepo(captureOverride, limit, prompt) {
     }
   };
   try {
-    const exitCode = await new Promise((resolve5, reject) => {
+    const exitCode = await new Promise((resolve6, reject) => {
       const child = spawn3(
         CLI2,
         [
@@ -49070,7 +49132,7 @@ async function runClaudeAnalyzeRepo(captureOverride, limit, prompt) {
           );
           return;
         }
-        resolve5(code2);
+        resolve6(code2);
       });
     });
     flushRedactors();
@@ -49910,7 +49972,7 @@ async function runCodexAnalyzeRepo(captureOverride, limit, prompt) {
     }
   };
   try {
-    const exitCode = await new Promise((resolve5, reject) => {
+    const exitCode = await new Promise((resolve6, reject) => {
       const child = spawn4(
         CLI3,
         [
@@ -49952,7 +50014,7 @@ async function runCodexAnalyzeRepo(captureOverride, limit, prompt) {
           );
           return;
         }
-        resolve5(code2);
+        resolve6(code2);
       });
     });
     flushRedactors();
@@ -50231,7 +50293,7 @@ async function runCursorAnalyzeRepo(captureOverride, limit, prompt) {
     }
   };
   try {
-    const exitCode = await new Promise((resolve5, reject) => {
+    const exitCode = await new Promise((resolve6, reject) => {
       const child = spawn5(
         "cursor",
         [
@@ -50276,7 +50338,7 @@ async function runCursorAnalyzeRepo(captureOverride, limit, prompt) {
           );
           return;
         }
-        resolve5(code2);
+        resolve6(code2);
       });
     });
     flushRedactors();
@@ -50790,7 +50852,7 @@ import * as p8 from "@clack/prompts";
 // src/setupV2Verification.ts
 import { spawnSync as spawnSync4 } from "child_process";
 import { existsSync as existsSync3, readFileSync as readFileSync4 } from "fs";
-import { basename, relative, resolve as resolve4 } from "path";
+import { basename, relative, resolve as resolve5 } from "path";
 var MUTATING_TOOLS = /* @__PURE__ */ new Set(["Edit", "Write"]);
 var CODE_EXTENSIONS = /* @__PURE__ */ new Set([
   ".cjs",
@@ -50814,7 +50876,7 @@ function pathFromToolInput(input2) {
   return typeof path15 === "string" ? path15 : void 0;
 }
 function normalizePath(cwd, path15) {
-  return resolve4(cwd, path15);
+  return resolve5(cwd, path15);
 }
 function readGitDirtyPaths(cwd) {
   const result = spawnSync4(
@@ -50915,7 +50977,7 @@ function resolveRelativeImport(registryPath, specifier, sourcePaths) {
   if (!specifier.startsWith(".")) {
     return void 0;
   }
-  const base = resolve4(registryPath, "..", specifier);
+  const base = resolve5(registryPath, "..", specifier);
   const candidates = [
     base,
     `${base}.ts`,
@@ -50924,8 +50986,8 @@ function resolveRelativeImport(registryPath, specifier, sourcePaths) {
     `${base}.mjs`,
     base.replace(/\.js$/, ".ts"),
     base.replace(/\.mjs$/, ".ts"),
-    resolve4(base, "index.ts"),
-    resolve4(base, "index.tsx")
+    resolve5(base, "index.ts"),
+    resolve5(base, "index.tsx")
   ];
   return candidates.find((candidate) => sourcePaths.has(candidate));
 }
@@ -50945,7 +51007,7 @@ function importedSourcePaths(registryPath, source, sourcePaths) {
   return [...paths];
 }
 function projectTypecheckCoversRegistry(cwd, registryPath) {
-  const configPath = resolve4(cwd, "tsconfig.json");
+  const configPath = resolve5(cwd, "tsconfig.json");
   if (!existsSync3(configPath)) {
     return false;
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bitfab-cli",
-  "version": "0.2.439",
+  "version": "0.2.440",
   "description": "Install and configure the Bitfab plugin in Claude Code, Codex, or Cursor, or the Bitfab skill pack in Amp (Alpha).",
   "type": "module",
   "main": "dist/index.js",