@tangle-network/agent-eval 0.174.0 → 0.175.0

package/dist/{opencode-sqlite-eK6HW6dr.js → claude-jsonl-CxZZrDJ3.js}

@@ -1,4 +1,3 @@
-import { createRequire } from "node:module";
 import { join } from "node:path";
 import { homedir } from "node:os";
 import { readFile, readdir } from "node:fs/promises";
@@ -28,7 +27,7 @@ async function findClaudeTranscripts(cwd, projectsDir = DEFAULT_CLAUDE_PROJECTS_
 		path: join(dir, n)
 	}));
 }
-const isRecord$1 = (v) => typeof v === "object" && v !== null && !Array.isArray(v);
+const isRecord = (v) => typeof v === "object" && v !== null && !Array.isArray(v);
 /** Parse transcript jsonl text into conversation lines. Non-conversation lines are dropped. */
 function parseClaudeEntries(raw) {
 	const out = [];
@@ -37,14 +36,14 @@ function parseClaudeEntries(raw) {
 		let entry;
 		try {
 			const parsed = JSON.parse(line);
-			if (!isRecord$1(parsed)) continue;
+			if (!isRecord(parsed)) continue;
 			entry = parsed;
 		} catch {
 			continue;
 		}
 		if (entry.type !== "user" && entry.type !== "assistant") continue;
 		const message = entry.message;
-		if (!isRecord$1(message)) continue;
+		if (!isRecord(message)) continue;
 		out.push({
 			type: entry.type,
 			timestamp: typeof entry.timestamp === "string" ? entry.timestamp : null,
@@ -59,7 +58,7 @@ function parseClaudeEntries(raw) {
 function blockText(content) {
 	if (typeof content === "string") return content;
 	if (!Array.isArray(content)) return "";
-	return content.filter((b) => isRecord$1(b) && b.type === "text" && typeof b.text === "string").map((b) => b.text).join("\n");
+	return content.filter((b) => isRecord(b) && b.type === "text" && typeof b.text === "string").map((b) => b.text).join("\n");
 }
 /** Parse one transcript jsonl into canonical messages + usage totals. */
 async function readClaudeTranscript(path, options = {}) {
@@ -101,7 +100,7 @@ function transcriptFromEntries(entries, options = {}) {
 			if (!Array.isArray(content)) continue;
 			let userText = "";
 			for (const block of content) {
-				if (!isRecord$1(block)) continue;
+				if (!isRecord(block)) continue;
 				if (block.type === "tool_result" && typeof block.tool_use_id === "string") messages.push({
 					role: "tool",
 					tool_call_id: block.tool_use_id,
@@ -119,7 +118,7 @@ function transcriptFromEntries(entries, options = {}) {
 		const apiId = typeof message.id === "string" ? message.id : null;
 		const continuesTurn = apiId !== null && apiId === lastAssistantApiId && lastAssistantIndex >= 0;
 		const msgUsage = message.usage;
-		if (isRecord$1(msgUsage) && !continuesTurn) {
+		if (isRecord(msgUsage) && !continuesTurn) {
 			usage.tokensIn += typeof msgUsage.input_tokens === "number" ? msgUsage.input_tokens : 0;
 			usage.tokensOut += typeof msgUsage.output_tokens === "number" ? msgUsage.output_tokens : 0;
 			usage.cacheRead += typeof msgUsage.cache_read_input_tokens === "number" ? msgUsage.cache_read_input_tokens : 0;
@@ -131,7 +130,7 @@ function transcriptFromEntries(entries, options = {}) {
 		let text = "";
 		const toolCalls = [];
 		for (const block of content) {
-			if (!isRecord$1(block)) continue;
+			if (!isRecord(block)) continue;
 			if (block.type === "thinking" && typeof block.thinking === "string" && block.thinking.length > 0) reasoning += (reasoning.length > 0 ? "\n" : "") + block.thinking;
 			else if (block.type === "text" && typeof block.text === "string") text += (text.length > 0 ? "\n" : "") + block.text;
 			else if (block.type === "tool_use" && typeof block.id === "string") toolCalls.push({
@@ -169,145 +168,6 @@ function transcriptFromEntries(entries, options = {}) {
 	};
 }
 //#endregion
-//#region src/rollout/readers/opencode-sqlite.ts
-/**
-* Read-only backfill reader over the opencode sqlite store
-* (~/.local/share/opencode/opencode.db) → canonical chat-with-tools messages.
-*
-* Schema consumed (observed, 2026-07): `session` rows carry directory /
-* parent_id / agent / model / cost / tokens_*; `message` rows carry a JSON
-* `data` blob ({role, modelID, providerID, tokens, cost, finish}); `part`
-* rows carry the actual content ({type: text|reasoning|tool|step-start|
-* step-finish|snapshot…}). Tool parts hold {callID, state:{input, output,
-* status}} — both the call and its result, which we split into an assistant
-* tool_call plus a role:"tool" result message.
-*
-* The store is mutable and can be corrupt (a `.corrupt-bak` sibling ships
-* next to it in the wild), so `openOpencodeDb` returns null instead of
-* throwing — callers record a gap line, never crash the backfill.
-*/
-const DEFAULT_OPENCODE_DB = join(homedir(), ".local", "share", "opencode", "opencode.db");
-/** Open the store read-only; null = unavailable/corrupt (caller records a gap). */
-async function openOpencodeDb(path = DEFAULT_OPENCODE_DB) {
-	try {
-		const { DatabaseSync } = createRequire(import.meta.url)("node:sqlite");
-		const db = new DatabaseSync(path, { readOnly: true });
-		db.prepare("SELECT id FROM session LIMIT 1").get();
-		return db;
-	} catch {
-		return null;
-	}
-}
-const isRecord = (v) => typeof v === "object" && v !== null && !Array.isArray(v);
-function parseSessionRow(row) {
-	let model = null;
-	if (typeof row.model === "string" && row.model.length > 0) try {
-		const parsed = JSON.parse(row.model);
-		if (isRecord(parsed)) model = parsed;
-	} catch {
-		model = null;
-	}
-	return {
-		id: String(row.id),
-		parentId: row.parent_id === null || row.parent_id === void 0 ? null : String(row.parent_id),
-		directory: String(row.directory),
-		agent: row.agent === null || row.agent === void 0 ? null : String(row.agent),
-		model,
-		costUsd: Number(row.cost ?? 0),
-		tokensInput: Number(row.tokens_input ?? 0),
-		tokensOutput: Number(row.tokens_output ?? 0),
-		tokensReasoning: Number(row.tokens_reasoning ?? 0),
-		tokensCacheRead: Number(row.tokens_cache_read ?? 0),
-		tokensCacheWrite: Number(row.tokens_cache_write ?? 0),
-		timeCreated: Number(row.time_created ?? 0),
-		timeUpdated: Number(row.time_updated ?? 0)
-	};
-}
-const SESSION_COLUMNS = "id, parent_id, directory, agent, model, cost, tokens_input, tokens_output, tokens_reasoning, tokens_cache_read, tokens_cache_write, time_created, time_updated";
-/** Sessions whose cwd is `directory` (the worker-clone join key). */
-function findOpencodeSessionsByDirectory(db, directory) {
-	return db.prepare(`SELECT ${SESSION_COLUMNS} FROM session WHERE directory = ? ORDER BY time_created`).all(directory).map(parseSessionRow);
-}
-function toolResultContent(output) {
-	if (typeof output === "string") return output;
-	if (output === null || output === void 0) return "";
-	return JSON.stringify(output);
-}
-/**
-* Convert one session's message+part rows into canonical messages.
-* An opencode assistant message row spans several model steps; each step's
-* parts (reasoning → text → tool …) become one assistant message followed by
-* the role:"tool" results of its calls, preserving order.
-*/
-function readOpencodeSessionMessages(db, sessionId) {
-	const messageRows = db.prepare("SELECT id, data FROM message WHERE session_id = ? ORDER BY time_created, id").all(sessionId);
-	const partsStmt = db.prepare("SELECT data FROM part WHERE message_id = ? ORDER BY id");
-	const messages = [];
-	for (const messageRow of messageRows) {
-		let data;
-		try {
-			const parsed = JSON.parse(messageRow.data);
-			if (!isRecord(parsed)) continue;
-			data = parsed;
-		} catch {
-			continue;
-		}
-		const parts = [];
-		for (const row of partsStmt.all(messageRow.id)) try {
-			const parsed = JSON.parse(row.data);
-			if (isRecord(parsed)) parts.push(parsed);
-		} catch {}
-		if (data.role === "user") {
-			const text = parts.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("\n");
-			messages.push({
-				role: "user",
-				content: text
-			});
-			continue;
-		}
-		if (data.role !== "assistant") continue;
-		const steps = [];
-		let current = [];
-		for (const part of parts) {
-			if (part.type === "step-start") {
-				if (current.length > 0) steps.push(current);
-				current = [];
-				continue;
-			}
-			if (part.type === "step-finish" || part.type === "snapshot" || part.type === "patch") continue;
-			current.push(part);
-		}
-		if (current.length > 0) steps.push(current);
-		for (const step of steps) {
-			const reasoning = step.filter((p) => p.type === "reasoning" && typeof p.text === "string" && p.text.length > 0).map((p) => p.text).join("\n");
-			const text = step.filter((p) => p.type === "text" && typeof p.text === "string").map((p) => p.text).join("\n");
-			const toolParts = step.filter((p) => p.type === "tool" && typeof p.callID === "string");
-			const toolCalls = toolParts.map((p) => ({
-				id: p.callID,
-				type: "function",
-				function: {
-					name: p.tool ?? "unknown",
-					arguments: JSON.stringify(p.state?.input ?? {})
-				}
-			}));
-			if (reasoning.length === 0 && text.length === 0 && toolCalls.length === 0) continue;
-			messages.push({
-				role: "assistant",
-				content: text.length > 0 ? text : null,
-				...reasoning.length > 0 ? { reasoning_content: reasoning } : {},
-				...toolCalls.length > 0 ? { tool_calls: toolCalls } : {}
-			});
-			for (const p of toolParts) messages.push({
-				role: "tool",
-				tool_call_id: p.callID,
-				name: p.tool ?? "unknown",
-				content: toolResultContent(p.state?.output)
-			});
-		}
-	}
-	return messages;
-}
-//#endregion
-export { DEFAULT_CLAUDE_PROJECTS_DIR as a, parseClaudeEntries as c, readOpencodeSessionMessages as i, readClaudeTranscript as l, findOpencodeSessionsByDirectory as n, claudeProjectSlug as o, openOpencodeDb as r, findClaudeTranscripts as s, DEFAULT_OPENCODE_DB as t, transcriptFromEntries as u };
+export { readClaudeTranscript as a, parseClaudeEntries as i, claudeProjectSlug as n, transcriptFromEntries as o, findClaudeTranscripts as r, DEFAULT_CLAUDE_PROJECTS_DIR as t };
 
-//# sourceMappingURL=opencode-sqlite-eK6HW6dr.js.map
+//# sourceMappingURL=claude-jsonl-CxZZrDJ3.js.map

package/dist/claude-jsonl-CxZZrDJ3.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"claude-jsonl-CxZZrDJ3.js","names":[],"sources":["../src/rollout/readers/claude-jsonl.ts"],"sourcesContent":["/**\n * Backfill reader over Claude Code project transcripts\n * (~/.claude/projects/<cwd-slug>/<sessionId>.jsonl) → canonical\n * chat-with-tools messages plus per-session token usage.\n *\n * Transcript lines consumed: type:\"user\" (string content or content blocks —\n * text + tool_result) and type:\"assistant\" (content blocks — thinking, text,\n * tool_use; message.usage carries tokens). Sidechain lines (isSidechain=true,\n * subagent threads) are separate invocations and are excluded from the main\n * transcript. Everything else (queue-operation, attachment, last-prompt…) is\n * transport metadata, not conversation.\n */\n\nimport { readdir, readFile } from 'node:fs/promises'\nimport { homedir } from 'node:os'\nimport { join } from 'node:path'\nimport type { ChatMessage, ChatToolCall } from '../schema'\n\nexport const DEFAULT_CLAUDE_PROJECTS_DIR = join(homedir(), '.claude', 'projects')\n\n/** Claude Code's project-directory slug for a working directory. */\nexport function claudeProjectSlug(cwd: string): string {\n  return cwd.replace(/[^a-zA-Z0-9-]/g, '-')\n}\n\nexport interface ClaudeTranscriptRef {\n  sessionId: string\n  path: string\n}\n\n/** Transcript files recorded for sessions launched from `cwd`. */\nexport async function findClaudeTranscripts(\n  cwd: string,\n  projectsDir: string = DEFAULT_CLAUDE_PROJECTS_DIR,\n): Promise<ClaudeTranscriptRef[]> {\n  const dir = join(projectsDir, claudeProjectSlug(cwd))\n  const names = await readdir(dir).catch(() => [])\n  return names\n    .filter((n) => n.endsWith('.jsonl'))\n    .sort()\n    .map((n) => ({ sessionId: n.replace(/\\.jsonl$/, ''), path: join(dir, n) }))\n}\n\nexport interface ClaudeUsageTotals {\n  tokensIn: number\n  tokensOut: number\n  cacheRead: number\n  cacheWrite: number\n}\n\nexport interface ClaudeTranscript {\n  messages: ChatMessage[]\n  usage: ClaudeUsageTotals\n  /** Timestamp of the first conversation line; null = empty transcript. */\n  startedAt: string | null\n  endedAt: string | null\n  model: string | null\n}\n\nconst isRecord = (v: unknown): v is Record<string, unknown> =>\n  typeof v === 'object' && v !== null && !Array.isArray(v)\n\n/**\n * One conversation line of a transcript, still in Claude Code's own shape.\n *\n * This is the single line-level parse of the format. `readClaudeTranscript`\n * projects it to canonical messages + usage; the supervision-tree reader\n * (`src/supervisor-run/claude-code-reader.ts`) projects the SAME entries to\n * spawn/settle/steer instants. Two projections, one parser — a second\n * transcript parser is how the two views silently disagree.\n */\nexport interface ClaudeEntry {\n  readonly type: 'user' | 'assistant'\n  /** ISO instant of the line; null when the line carried none. */\n  readonly timestamp: string | null\n  /** The Anthropic message body (`role`, `content`, `model`, `usage`). */\n  readonly message: Record<string, unknown>\n  /** Claude Code's structured tool result, when the line carries one. */\n  readonly toolUseResult: unknown\n  /** True on subagent threads — a separate invocation, not this transcript's turn. */\n  readonly isSidechain: boolean\n  /** Subagent id Claude Code stamps on sidechain lines; null on main-thread lines. */\n  readonly agentId: string | null\n}\n\n/** Parse transcript jsonl text into conversation lines. Non-conversation lines are dropped. */\nexport function parseClaudeEntries(raw: string): ClaudeEntry[] {\n  const out: ClaudeEntry[] = []\n  for (const line of raw.split('\\n')) {\n    if (!line.trim()) continue\n    let entry: Record<string, unknown>\n    try {\n      const parsed: unknown = JSON.parse(line)\n      if (!isRecord(parsed)) continue\n      entry = parsed\n    } catch {\n      continue\n    }\n    if (entry.type !== 'user' && entry.type !== 'assistant') continue\n    const message = entry.message\n    if (!isRecord(message)) continue\n    out.push({\n      type: entry.type,\n      timestamp: typeof entry.timestamp === 'string' ? entry.timestamp : null,\n      message,\n      toolUseResult: entry.toolUseResult,\n      isSidechain: entry.isSidechain === true,\n      agentId: typeof entry.agentId === 'string' ? entry.agentId : null,\n    })\n  }\n  return out\n}\n\nexport interface ReadClaudeTranscriptOptions {\n  /**\n   * Read the sidechain (subagent) thread instead of skipping it. Subagent\n   * transcripts under `<session>/subagents/agent-<id>.jsonl` are sidechain\n   * lines end to end, so their usage is invisible without this.\n   */\n  readonly includeSidechain?: boolean\n}\n\nfunction blockText(content: unknown): string {\n  if (typeof content === 'string') return content\n  if (!Array.isArray(content)) return ''\n  return content\n    .filter(\n      (b): b is Record<string, unknown> =>\n        isRecord(b) && b.type === 'text' && typeof b.text === 'string',\n    )\n    .map((b) => b.text as string)\n    .join('\\n')\n}\n\n/** Parse one transcript jsonl into canonical messages + usage totals. */\nexport async function readClaudeTranscript(\n  path: string,\n  options: ReadClaudeTranscriptOptions = {},\n): Promise<ClaudeTranscript> {\n  return transcriptFromEntries(parseClaudeEntries(await readFile(path, 'utf8')), options)\n}\n\n/** The messages+usage projection of already-parsed entries. */\nexport function transcriptFromEntries(\n  entries: readonly ClaudeEntry[],\n  options: ReadClaudeTranscriptOptions = {},\n): ClaudeTranscript {\n  const wantSidechain = options.includeSidechain === true\n  const messages: ChatMessage[] = []\n  const usage: ClaudeUsageTotals = { tokensIn: 0, tokensOut: 0, cacheRead: 0, cacheWrite: 0 }\n  let startedAt: string | null = null\n  let endedAt: string | null = null\n  let model: string | null = null\n  // Claude Code writes one jsonl line PER CONTENT BLOCK of an API message,\n  // repeating message.id and usage on each — merge blocks into one canonical\n  // assistant turn and count usage once per API message id.\n  let lastAssistantApiId: string | null = null\n  let lastAssistantIndex = -1\n\n  for (const entry of entries) {\n    if (entry.isSidechain !== wantSidechain) continue\n    const message = entry.message\n    if (entry.timestamp !== null) {\n      if (startedAt === null) startedAt = entry.timestamp\n      endedAt = entry.timestamp\n    }\n\n    if (entry.type === 'user') {\n      lastAssistantApiId = null\n      lastAssistantIndex = -1\n      const content = message.content\n      if (typeof content === 'string') {\n        messages.push({ role: 'user', content })\n        continue\n      }\n      if (!Array.isArray(content)) continue\n      // A user line may interleave tool_result blocks (answers to the prior\n      // assistant tool_use) with plain text; preserve order.\n      let userText = ''\n      for (const block of content) {\n        if (!isRecord(block)) continue\n        if (block.type === 'tool_result' && typeof block.tool_use_id === 'string') {\n          messages.push({\n            role: 'tool',\n            tool_call_id: block.tool_use_id,\n            content:\n              blockText(block.content) || (typeof block.content === 'string' ? block.content : ''),\n          })\n        } else if (block.type === 'text' && typeof block.text === 'string') {\n          userText += (userText.length > 0 ? '\\n' : '') + block.text\n        }\n      }\n      if (userText.length > 0) messages.push({ role: 'user', content: userText })\n      continue\n    }\n\n    // assistant\n    if (typeof message.model === 'string') model = message.model\n    const apiId = typeof message.id === 'string' ? message.id : null\n    const continuesTurn = apiId !== null && apiId === lastAssistantApiId && lastAssistantIndex >= 0\n    const msgUsage = message.usage\n    if (isRecord(msgUsage) && !continuesTurn) {\n      usage.tokensIn += typeof msgUsage.input_tokens === 'number' ? msgUsage.input_tokens : 0\n      usage.tokensOut += typeof msgUsage.output_tokens === 'number' ? msgUsage.output_tokens : 0\n      usage.cacheRead +=\n        typeof msgUsage.cache_read_input_tokens === 'number' ? msgUsage.cache_read_input_tokens : 0\n      usage.cacheWrite +=\n        typeof msgUsage.cache_creation_input_tokens === 'number'\n          ? msgUsage.cache_creation_input_tokens\n          : 0\n    }\n    const content = message.content\n    if (!Array.isArray(content)) continue\n    let reasoning = ''\n    let text = ''\n    const toolCalls: ChatToolCall[] = []\n    for (const block of content) {\n      if (!isRecord(block)) continue\n      if (\n        block.type === 'thinking' &&\n        typeof block.thinking === 'string' &&\n        block.thinking.length > 0\n      ) {\n        reasoning += (reasoning.length > 0 ? '\\n' : '') + block.thinking\n      } else if (block.type === 'text' && typeof block.text === 'string') {\n        text += (text.length > 0 ? '\\n' : '') + block.text\n      } else if (block.type === 'tool_use' && typeof block.id === 'string') {\n        toolCalls.push({\n          id: block.id,\n          type: 'function',\n          function: {\n            name: typeof block.name === 'string' ? block.name : 'unknown',\n            arguments: JSON.stringify(block.input ?? {}),\n          },\n        })\n      }\n    }\n    if (reasoning.length === 0 && text.length === 0 && toolCalls.length === 0) continue\n    if (continuesTurn) {\n      const prev = messages[lastAssistantIndex]!\n      if (text.length > 0) prev.content = prev.content === null ? text : `${prev.content}\\n${text}`\n      if (reasoning.length > 0) {\n        prev.reasoning_content =\n          prev.reasoning_content === undefined\n            ? reasoning\n            : `${prev.reasoning_content}\\n${reasoning}`\n      }\n      if (toolCalls.length > 0) prev.tool_calls = [...(prev.tool_calls ?? []), ...toolCalls]\n      continue\n    }\n    messages.push({\n      role: 'assistant',\n      content: text.length > 0 ? text : null,\n      ...(reasoning.length > 0 ? { reasoning_content: reasoning } : {}),\n      ...(toolCalls.length > 0 ? { tool_calls: toolCalls } : {}),\n    })\n    lastAssistantApiId = apiId\n    lastAssistantIndex = messages.length - 1\n  }\n\n  return { messages, usage, startedAt, endedAt, model }\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAkBA,MAAa,8BAA8B,KAAK,QAAQ,GAAG,WAAW,UAAU;;AAGhF,SAAgB,kBAAkB,KAAqB;CACrD,OAAO,IAAI,QAAQ,kBAAkB,GAAG;AAC1C;;AAQA,eAAsB,sBACpB,KACA,cAAsB,6BACU;CAChC,MAAM,MAAM,KAAK,aAAa,kBAAkB,GAAG,CAAC;CAEpD,QAAO,MADa,QAAQ,GAAG,CAAC,CAAC,YAAY,CAAC,CAAC,EAAA,CAE5C,QAAQ,MAAM,EAAE,SAAS,QAAQ,CAAC,CAAC,CACnC,KAAK,CAAC,CACN,KAAK,OAAO;EAAE,WAAW,EAAE,QAAQ,YAAY,EAAE;EAAG,MAAM,KAAK,KAAK,CAAC;CAAE,EAAE;AAC9E;AAkBA,MAAM,YAAY,MAChB,OAAO,MAAM,YAAY,MAAM,QAAQ,CAAC,MAAM,QAAQ,CAAC;;AA0BzD,SAAgB,mBAAmB,KAA4B;CAC7D,MAAM,MAAqB,CAAC;CAC5B,KAAK,MAAM,QAAQ,IAAI,MAAM,IAAI,GAAG;EAClC,IAAI,CAAC,KAAK,KAAK,GAAG;EAClB,IAAI;EACJ,IAAI;GACF,MAAM,SAAkB,KAAK,MAAM,IAAI;GACvC,IAAI,CAAC,SAAS,MAAM,GAAG;GACvB,QAAQ;EACV,QAAQ;GACN;EACF;EACA,IAAI,MAAM,SAAS,UAAU,MAAM,SAAS,aAAa;EACzD,MAAM,UAAU,MAAM;EACtB,IAAI,CAAC,SAAS,OAAO,GAAG;EACxB,IAAI,KAAK;GACP,MAAM,MAAM;GACZ,WAAW,OAAO,MAAM,cAAc,WAAW,MAAM,YAAY;GACnE;GACA,eAAe,MAAM;GACrB,aAAa,MAAM,gBAAgB;GACnC,SAAS,OAAO,MAAM,YAAY,WAAW,MAAM,UAAU;EAC/D,CAAC;CACH;CACA,OAAO;AACT;AAWA,SAAS,UAAU,SAA0B;CAC3C,IAAI,OAAO,YAAY,UAAU,OAAO;CACxC,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG,OAAO;CACpC,OAAO,QACJ,QACE,MACC,SAAS,CAAC,KAAK,EAAE,SAAS,UAAU,OAAO,EAAE,SAAS,QAC1D,CAAC,CACA,KAAK,MAAM,EAAE,IAAc,CAAC,CAC5B,KAAK,IAAI;AACd;;AAGA,eAAsB,qBACpB,MACA,UAAuC,CAAC,GACb;CAC3B,OAAO,sBAAsB,mBAAmB,MAAM,SAAS,MAAM,MAAM,CAAC,GAAG,OAAO;AACxF;;AAGA,SAAgB,sBACd,SACA,UAAuC,CAAC,GACtB;CAClB,MAAM,gBAAgB,QAAQ,qBAAqB;CACnD,MAAM,WAA0B,CAAC;CACjC,MAAM,QAA2B;EAAE,UAAU;EAAG,WAAW;EAAG,WAAW;EAAG,YAAY;CAAE;CAC1F,IAAI,YAA2B;CAC/B,IAAI,UAAyB;CAC7B,IAAI,QAAuB;CAI3B,IAAI,qBAAoC;CACxC,IAAI,qBAAqB;CAEzB,KAAK,MAAM,SAAS,SAAS;EAC3B,IAAI,MAAM,gBAAgB,eAAe;EACzC,MAAM,UAAU,MAAM;EACtB,IAAI,MAAM,cAAc,MAAM;GAC5B,IAAI,cAAc,MAAM,YAAY,MAAM;GAC1C,UAAU,MAAM;EAClB;EAEA,IAAI,MAAM,SAAS,QAAQ;GACzB,qBAAqB;GACrB,qBAAqB;GACrB,MAAM,UAAU,QAAQ;GACxB,IAAI,OAAO,YAAY,UAAU;IAC/B,SAAS,KAAK;KAAE,MAAM;KAAQ;IAAQ,CAAC;IACvC;GACF;GACA,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG;GAG7B,IAAI,WAAW;GACf,KAAK,MAAM,SAAS,SAAS;IAC3B,IAAI,CAAC,SAAS,KAAK,GAAG;IACtB,IAAI,MAAM,SAAS,iBAAiB,OAAO,MAAM,gBAAgB,UAC/D,SAAS,KAAK;KACZ,MAAM;KACN,cAAc,MAAM;KACpB,SACE,UAAU,MAAM,OAAO,MAAM,OAAO,MAAM,YAAY,WAAW,MAAM,UAAU;IACrF,CAAC;SACI,IAAI,MAAM,SAAS,UAAU,OAAO,MAAM,SAAS,UACxD,aAAa,SAAS,SAAS,IAAI,OAAO,MAAM,MAAM;GAE1D;GACA,IAAI,SAAS,SAAS,GAAG,SAAS,KAAK;IAAE,MAAM;IAAQ,SAAS;GAAS,CAAC;GAC1E;EACF;EAGA,IAAI,OAAO,QAAQ,UAAU,UAAU,QAAQ,QAAQ;EACvD,MAAM,QAAQ,OAAO,QAAQ,OAAO,WAAW,QAAQ,KAAK;EAC5D,MAAM,gBAAgB,UAAU,QAAQ,UAAU,sBAAsB,sBAAsB;EAC9F,MAAM,WAAW,QAAQ;EACzB,IAAI,SAAS,QAAQ,KAAK,CAAC,eAAe;GACxC,MAAM,YAAY,OAAO,SAAS,iBAAiB,WAAW,SAAS,eAAe;GACtF,MAAM,aAAa,OAAO,SAAS,kBAAkB,WAAW,SAAS,gBAAgB;GACzF,MAAM,aACJ,OAAO,SAAS,4BAA4B,WAAW,SAAS,0BAA0B;GAC5F,MAAM,cACJ,OAAO,SAAS,gCAAgC,WAC5C,SAAS,8BACT;EACR;EACA,MAAM,UAAU,QAAQ;EACxB,IAAI,CAAC,MAAM,QAAQ,OAAO,GAAG;EAC7B,IAAI,YAAY;EAChB,IAAI,OAAO;EACX,MAAM,YAA4B,CAAC;EACnC,KAAK,MAAM,SAAS,SAAS;GAC3B,IAAI,CAAC,SAAS,KAAK,GAAG;GACtB,IACE,MAAM,SAAS,cACf,OAAO,MAAM,aAAa,YAC1B,MAAM,SAAS,SAAS,GAExB,cAAc,UAAU,SAAS,IAAI,OAAO,MAAM,MAAM;QACnD,IAAI,MAAM,SAAS,UAAU,OAAO,MAAM,SAAS,UACxD,SAAS,KAAK,SAAS,IAAI,OAAO,MAAM,MAAM;QACzC,IAAI,MAAM,SAAS,cAAc,OAAO,MAAM,OAAO,UAC1D,UAAU,KAAK;IACb,IAAI,MAAM;IACV,MAAM;IACN,UAAU;KACR,MAAM,OAAO,MAAM,SAAS,WAAW,MAAM,OAAO;KACpD,WAAW,KAAK,UAAU,MAAM,SAAS,CAAC,CAAC;IAC7C;GACF,CAAC;EAEL;EACA,IAAI,UAAU,WAAW,KAAK,KAAK,WAAW,KAAK,UAAU,WAAW,GAAG;EAC3E,IAAI,eAAe;GACjB,MAAM,OAAO,SAAS;GACtB,IAAI,KAAK,SAAS,GAAG,KAAK,UAAU,KAAK,YAAY,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI;GACvF,IAAI,UAAU,SAAS,GACrB,KAAK,oBACH,KAAK,sBAAsB,KAAA,IACvB,YACA,GAAG,KAAK,kBAAkB,IAAI;GAEtC,IAAI,UAAU,SAAS,GAAG,KAAK,aAAa,CAAC,GAAI,KAAK,cAAc,CAAC,GAAI,GAAG,SAAS;GACrF;EACF;EACA,SAAS,KAAK;GACZ,MAAM;GACN,SAAS,KAAK,SAAS,IAAI,OAAO;GAClC,GAAI,UAAU,SAAS,IAAI,EAAE,mBAAmB,UAAU,IAAI,CAAC;GAC/D,GAAI,UAAU,SAAS,IAAI,EAAE,YAAY,UAAU,IAAI,CAAC;EAC1D,CAAC;EACD,qBAAqB;EACrB,qBAAqB,SAAS,SAAS;CACzC;CAEA,OAAO;EAAE;EAAU;EAAO;EAAW;EAAS;CAAM;AACtD"}

package/dist/cli.js

@@ -1,8 +1,9 @@
 #!/usr/bin/env node
 import { i as runRolloutReleaseCli } from "./hf-dataset-D8_RNIis.js";
 import { t as createChatClient } from "./chat-client-CkmjYlfB.js";
-import { t as runAnalystBenchmarkCommand } from "./benchmark-command-mZIlR-ra.js";
+import { t as runAnalystBenchmarkCommand } from "./benchmark-command-D_5xG9LG.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-D_cjseFN.js";
+import { t as runSupervisorRunCommand } from "./report-command-DKlXfU5r.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts
 /**
@@ -84,6 +85,7 @@ function nonEmpty(value) {
 *   agent-eval rpc <method>          # one request from stdin → one response on stdout
 *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout
 *   agent-eval analyst-benchmark ... # benchmark a real-model trace analyst
+*   agent-eval supervisor-run report <runDir> [--format headline|markdown|json]
 *   agent-eval openapi [--out path]  # write OpenAPI spec
 *   agent-eval version
 *
@@ -140,6 +142,10 @@ Commands:
   analyst-benchmark --dataset <agentrx|codetracebench> --labels <path> --trace-dir <path> [--artifact-dir <path>] --out <dir> ...
         Compare an empty baseline with a real-model trace analyst on public labels.
         Run with --help for all required inputs and controls.
+  supervisor-run report <runDir> [--format headline|markdown|json]
+        Report one Runtime or loops supervisor run directory. The status comes
+        from Runtime's own result.json / failure.json; every missing measurement
+        stays named. Exits 1 when the directory cannot be read.
   version
         Print server + wire-protocol version JSON.
 
@@ -204,6 +210,7 @@ async function main() {
 		}
 		case "rollout-release": return await runRolloutReleaseCli(process.argv.slice(3));
 		case "analyst-benchmark": return await runAnalystBenchmarkCommand(process.argv.slice(3));
+		case "supervisor-run": return await runSupervisorRunCommand(process.argv.slice(3));
 		case "version":
 			process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\n`);
 			return 0;
@@ -238,7 +245,7 @@ const FLAGS_BY_COMMAND = {
 	"": /* @__PURE__ */ new Set()
 };
 function assertKnownFlags(command, flags) {
-	if (command === "rollout-release" || command === "analyst-benchmark") return;
+	if (command === "rollout-release" || command === "analyst-benchmark" || command === "supervisor-run") return;
 	const allowed = FLAGS_BY_COMMAND[command];
 	if (!allowed) return;
 	const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag));

package/dist/cli.js.map

@@ -1 +1 @@
-{"version":3,"file":"cli.js","names":[],"sources":["../src/cli-config.ts","../src/cli.ts"],"sourcesContent":["/**\n * Provider configuration for the `agent-eval` binary.\n *\n * This is the ONE place in the package that turns an environment credential\n * into a model transport, and it exists only inside the binary. The `agent-eval`\n * server is a deployed process whose caller is a JSON-RPC or HTTP client in\n * another language, so it cannot be handed a `ChatClient`; it reads its own\n * credential the way every server does. The library never does: a TypeScript\n * consumer binds its own transport and agent-eval holds no provider key.\n */\n\nimport { type ChatClient, createChatClient } from './analyst/chat-client'\n\nexport interface CliLlmConfig {\n  /** Judge transport for the wire handlers. Absent when no provider is configured. */\n  chat?: ChatClient\n  model?: string\n}\n\n/** The endpoint the binary resolved from its environment. */\nexport interface CliProviderRoute {\n  baseUrl: string\n  apiKey: string\n  model?: string\n}\n\n/**\n * Environment precedence for the binary's provider route, with no client built.\n *\n * Both halves are required: a base URL without a key, or a key without an\n * endpoint, is a half-configured server. `/v1/judge` then refuses with\n * `llm_not_configured` instead of calling an unintended endpoint.\n */\nexport function resolveCliProviderRoute(\n  env: NodeJS.ProcessEnv = process.env,\n): CliProviderRoute | undefined {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  if (!baseUrl || !apiKey) return undefined\n  return { baseUrl, apiKey, ...(model ? { model } : {}) }\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const route = resolveCliProviderRoute(env)\n  if (!route) return {}\n  return {\n    chat: cliChatClient({ baseUrl: route.baseUrl, apiKey: route.apiKey }, route.model),\n    ...(route.model ? { model: route.model } : {}),\n  }\n}\n\n/**\n * The binary's transport IS the published one — `createChatClient({ transport:\n * 'openai-compatible' })` — so the server and a library consumer cannot drift\n * onto two different clients for the same endpoint.\n *\n * The one thing kept local is the error text. A generic \"no model on the\n * request\" tells a server operator nothing; this names the variable to set.\n */\nfunction cliChatClient(\n  route: { baseUrl: string; apiKey: string },\n  defaultModel: string | undefined,\n): ChatClient {\n  const client = createChatClient({\n    transport: 'openai-compatible',\n    baseUrl: route.baseUrl,\n    apiKey: route.apiKey,\n    ...(defaultModel ? { defaultModel } : {}),\n  })\n  return {\n    ...client,\n    chat: (req, callOpts) => {\n      if (!req.model && !defaultModel) {\n        throw new Error(\n          'agent-eval: no model on the request and no AGENT_EVAL_LLM_MODEL configured',\n        )\n      }\n      return client.chat(req, callOpts)\n    },\n  }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n","#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval analyst-benchmark ... # benchmark a real-model trace analyst\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { runAnalystBenchmarkCommand } from './analyst/benchmark-command'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Methods: judge, listRubrics, version.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n        validate, fail-closed split filter, scrub, export formats + card.\n  analyst-benchmark --dataset <agentrx|codetracebench> --labels <path> --trace-dir <path> [--artifact-dir <path>] --out <dir> ...\n        Compare an empty baseline with a real-model trace analyst on public labels.\n        Run with --help for all required inputs and controls.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted. This binary is the only\n  place that reads a provider credential; the library never does. Without both\n  a base URL and a key, /v1/judge refuses with llm_not_configured.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (command === 'analyst-benchmark' && flags.help === 'true') {\n    return await runAnalystBenchmarkCommand(process.argv.slice(3))\n  }\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'rollout-release': {\n      // The subcommand owns its own flag grammar (multi-value --formats,\n      // boolean --include-proposers) — pass raw argv through untouched.\n      return await runRolloutReleaseCli(process.argv.slice(3))\n    }\n    case 'analyst-benchmark': {\n      return await runAnalystBenchmarkCommand(process.argv.slice(3))\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  // These subcommands parse their own argv.\n  if (command === 'rollout-release' || command === 'analyst-benchmark') return\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;AAiCA,SAAgB,wBACd,MAAyB,QAAQ,KACH;CAC9B,MAAM,kBAAkB,SAAS,IAAI,uBAAuB;CAC5D,MAAM,iBAAiB,SAAS,IAAI,sBAAsB;CAC1D,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,KAAA,OAC7C,eAAe,mCAAmC,KAAA;CACrD,MAAM,SAAS,kBAAkB,gBAAgB;CACjD,MAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;CAE/F,IAAI,CAAC,WAAW,CAAC,QAAQ,OAAO,KAAA;CAChC,OAAO;EAAE;EAAS;EAAQ,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;CAAG;AACxD;AAEA,SAAgB,oBAAoB,MAAyB,QAAQ,KAAmB;CACtF,MAAM,QAAQ,wBAAwB,GAAG;CACzC,IAAI,CAAC,OAAO,OAAO,CAAC;CACpB,OAAO;EACL,MAAM,cAAc;GAAE,SAAS,MAAM;GAAS,QAAQ,MAAM;EAAO,GAAG,MAAM,KAAK;EACjF,GAAI,MAAM,QAAQ,EAAE,OAAO,MAAM,MAAM,IAAI,CAAC;CAC9C;AACF;;;;;;;;;AAUA,SAAS,cACP,OACA,cACY;CACZ,MAAM,SAAS,iBAAiB;EAC9B,WAAW;EACX,SAAS,MAAM;EACf,QAAQ,MAAM;EACd,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;CACzC,CAAC;CACD,OAAO;EACL,GAAG;EACH,OAAO,KAAK,aAAa;GACvB,IAAI,CAAC,IAAI,SAAS,CAAC,cACjB,MAAM,IAAI,MACR,4EACF;GAEF,OAAO,OAAO,KAAK,KAAK,QAAQ;EAClC;CACF;AACF;AAEA,SAAS,SAAS,OAA+C;CAC/D,MAAM,UAAU,OAAO,KAAK;CAC5B,OAAO,UAAU,UAAU,KAAA;AAC7B;;;;;;;;;;;;;;;;ACpEA,SAAS,UAAU,MAAsB;CACvC,MAAM,CAAC,SAAS,GAAG,QAAQ;CAC3B,MAAM,aAAuB,CAAC;CAC9B,MAAM,QAAgC,CAAC;CACvC,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,IAAI,IAAI,WAAW,IAAI,GAAG;GACxB,MAAM,MAAM,IAAI,MAAM,CAAC;GACvB,MAAM,WAAW,IAAI,QAAQ,GAAG;GAChC,IAAI,YAAY,GAAG;IACjB,MAAM,IAAI,MAAM,GAAG,QAAQ,KAAK,IAAI,MAAM,WAAW,CAAC;IACtD;GACF;GACA,MAAM,MAAM;GACZ,IAAI,QAAQ,QAAQ;IAClB,MAAM,OAAO;IACb;GACF;GACA,MAAM,OAAO,KAAK,IAAI;GACtB,IAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;IAC1C,MAAM,OAAO;IACb;GACF,OACE,MAAM,OAAO;EAEjB,OAAO,IAAI,QAAQ,MACjB,MAAM,OAAO;OAEb,WAAW,KAAK,GAAG;CAEvB;CACA,OAAO;EAAE,SAAS,WAAW;EAAQ;EAAY;CAAM;AACzD;AAEA,MAAM,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;AA6Bb,eAAe,OAAwB;CACrC,MAAM,EAAE,SAAS,YAAY,UAAU,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;CACtE,iBAAiB,SAAS,KAAK;CAE/B,IAAI,YAAY,uBAAuB,MAAM,SAAS,QACpD,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;CAE/D,IAAI,MAAM,SAAS,QAAQ;EACzB,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;EAChC,OAAO;CACT;CAEA,QAAQ,SAAR;EACE,KAAK,SAAS;GACZ,MAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;GAC3C,MAAM,OAAO,MAAM,QAAQ;GAC3B,MAAM,MAAM,oBAAoB;GAChC,MAAM,EAAE,WAAW,MAAM,iBAAiB;IACxC;IACA;IACA,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;GAED,MAAM,YAAY,QAAgB;IAEhC,QAAQ,IAAI,yBAAyB,IAAI,gBAAgB;IACzD,OAAO,YAAY,QAAQ,KAAK,CAAC,CAAC;IAElC,iBAAiB,QAAQ,KAAK,CAAC,GAAG,GAAI,CAAC,CAAC,MAAM;GAChD;GACA,QAAQ,GAAG,gBAAgB,SAAS,QAAQ,CAAC;GAC7C,QAAQ,GAAG,iBAAiB,SAAS,SAAS,CAAC;GAE/C,MAAM,IAAI,cAAc,CAAC,CAAC;GAC1B,OAAO;EACT;EACA,KAAK,OAAO;GACV,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,WAAW,QAAQ;IAC9B,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;EACH;EACA,KAAK,aAAa;GAChB,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,YAAY,QAAQ;IAC/B,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;EACH;EACA,KAAK,WAAW;GACd,MAAM,MAAM,MAAM,OAAO;GACzB,MAAM,OAAO,aAAa,cAAc,CAAC,CAAC,OAAO;GACjD,cAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,EAAE,KAAK,OAAO;GAEhE,QAAQ,IAAI,0CAA0C,KAAK;GAC3D,OAAO;EACT;EACA,KAAK,mBAGH,OAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAEzD,KAAK,qBACH,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;EAE/D,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,EAAE,GAAG;GACpE,OAAO;EAET,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,cAAc,CAAC,CAAC,QAAQ,GAAG;GACnD,OAAO;EAET,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;GAChC,OAAO;EACT;GACE,QAAQ,OAAO,MAAM,oBAAoB,QAAQ,IAAI,KAAK,GAAG;GAC7D,OAAO;CACX;AACF;AAEA,MAAM,mBAAwD;CAC5D,uBAAO,IAAI,IAAI;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,qBAAK,IAAI,IAAI,CAAC,MAAM,CAAC;CACrB,6BAAa,IAAI,IAAI,CAAC,MAAM,CAAC;CAC7B,yBAAS,IAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;CAChC,yBAAS,IAAI,IAAI,CAAC,MAAM,CAAC;CACzB,sBAAM,IAAI,IAAI;CACd,0BAAU,IAAI,IAAI;CAClB,sBAAM,IAAI,IAAI;CACd,6BAAa,IAAI,IAAI;CACrB,oBAAI,IAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;CAE9E,IAAI,YAAY,qBAAqB,YAAY,qBAAqB;CACtE,MAAM,UAAU,iBAAiB;CACjC,IAAI,CAAC,SAAS;CACd,MAAM,UAAU,OAAO,KAAK,KAAK,CAAC,CAAC,QAAQ,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;CACtE,IAAI,QAAQ,SAAS,GACnB,MAAM,IAAI,MAAM,oBAAoB,WAAW,OAAO,MAAM,QAAQ,IAAI;AAE5E;AAEA,SAAS,UAAU,KAAqB;CACtC,MAAM,OAAO,OAAO,GAAG;CACvB,IAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAChD,MAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,GAAG;CAE9F,OAAO;AACT;AAEA,KAAK,CAAC,CACH,MAAM,SAAS,QAAQ,KAAK,IAAI,CAAC,CAAC,CAClC,OAAO,QAAQ;CAEd,QAAQ,MAAM,2BAA2B,GAAG;CAC5C,QAAQ,KAAK,CAAC;AAChB,CAAC"}
+{"version":3,"file":"cli.js","names":[],"sources":["../src/cli-config.ts","../src/cli.ts"],"sourcesContent":["/**\n * Provider configuration for the `agent-eval` binary.\n *\n * This is the ONE place in the package that turns an environment credential\n * into a model transport, and it exists only inside the binary. The `agent-eval`\n * server is a deployed process whose caller is a JSON-RPC or HTTP client in\n * another language, so it cannot be handed a `ChatClient`; it reads its own\n * credential the way every server does. The library never does: a TypeScript\n * consumer binds its own transport and agent-eval holds no provider key.\n */\n\nimport { type ChatClient, createChatClient } from './analyst/chat-client'\n\nexport interface CliLlmConfig {\n  /** Judge transport for the wire handlers. Absent when no provider is configured. */\n  chat?: ChatClient\n  model?: string\n}\n\n/** The endpoint the binary resolved from its environment. */\nexport interface CliProviderRoute {\n  baseUrl: string\n  apiKey: string\n  model?: string\n}\n\n/**\n * Environment precedence for the binary's provider route, with no client built.\n *\n * Both halves are required: a base URL without a key, or a key without an\n * endpoint, is a half-configured server. `/v1/judge` then refuses with\n * `llm_not_configured` instead of calling an unintended endpoint.\n */\nexport function resolveCliProviderRoute(\n  env: NodeJS.ProcessEnv = process.env,\n): CliProviderRoute | undefined {\n  const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n  const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n  const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n  const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n  const baseUrl =\n    explicitBaseUrl ??\n    nonEmpty(env.OPENAI_BASE_URL) ??\n    nonEmpty(env.TANGLE_ROUTER_URL) ??\n    (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n    (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n  const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n  const model =\n    nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n  if (!baseUrl || !apiKey) return undefined\n  return { baseUrl, apiKey, ...(model ? { model } : {}) }\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n  const route = resolveCliProviderRoute(env)\n  if (!route) return {}\n  return {\n    chat: cliChatClient({ baseUrl: route.baseUrl, apiKey: route.apiKey }, route.model),\n    ...(route.model ? { model: route.model } : {}),\n  }\n}\n\n/**\n * The binary's transport IS the published one — `createChatClient({ transport:\n * 'openai-compatible' })` — so the server and a library consumer cannot drift\n * onto two different clients for the same endpoint.\n *\n * The one thing kept local is the error text. A generic \"no model on the\n * request\" tells a server operator nothing; this names the variable to set.\n */\nfunction cliChatClient(\n  route: { baseUrl: string; apiKey: string },\n  defaultModel: string | undefined,\n): ChatClient {\n  const client = createChatClient({\n    transport: 'openai-compatible',\n    baseUrl: route.baseUrl,\n    apiKey: route.apiKey,\n    ...(defaultModel ? { defaultModel } : {}),\n  })\n  return {\n    ...client,\n    chat: (req, callOpts) => {\n      if (!req.model && !defaultModel) {\n        throw new Error(\n          'agent-eval: no model on the request and no AGENT_EVAL_LLM_MODEL configured',\n        )\n      }\n      return client.chat(req, callOpts)\n    },\n  }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n  const trimmed = value?.trim()\n  return trimmed ? trimmed : undefined\n}\n","#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n *   agent-eval serve [--port 5005] [--host 127.0.0.1]\n *   agent-eval rpc <method>          # one request from stdin → one response on stdout\n *   agent-eval rpc-batch <method>    # JSONL stdin → JSONL stdout\n *   agent-eval analyst-benchmark ... # benchmark a real-model trace analyst\n *   agent-eval supervisor-run report <runDir> [--format headline|markdown|json]\n *   agent-eval openapi [--out path]  # write OpenAPI spec\n *   agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { runAnalystBenchmarkCommand } from './analyst/benchmark-command'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { runSupervisorRunCommand } from './supervisor-run/report-command'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n  command: string\n  positional: string[]\n  flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n  const [command, ...rest] = argv\n  const positional: string[] = []\n  const flags: Record<string, string> = {}\n  for (let i = 0; i < rest.length; i++) {\n    const tok = rest[i]!\n    if (tok.startsWith('--')) {\n      const raw = tok.slice(2)\n      const equalsAt = raw.indexOf('=')\n      if (equalsAt >= 0) {\n        flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n        continue\n      }\n      const key = raw\n      if (key === 'help') {\n        flags[key] = 'true'\n        continue\n      }\n      const next = rest[i + 1]\n      if (next != null && !next.startsWith('--')) {\n        flags[key] = next\n        i++\n      } else {\n        flags[key] = 'true'\n      }\n    } else if (tok === '-h') {\n      flags.help = 'true'\n    } else {\n      positional.push(tok)\n    }\n  }\n  return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n  serve [--port 5005] [--host 127.0.0.1]\n        Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n  rpc <method>\n        Read one JSON object from stdin (the params for <method>), write one\n        JSON object to stdout. Methods: judge, listRubrics, version.\n  rpc-batch <method>\n        Like 'rpc' but JSONL in / JSONL out.\n  openapi [--out openapi.json]\n        Write the OpenAPI 3.1 spec.\n  rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n        Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n        validate, fail-closed split filter, scrub, export formats + card.\n  analyst-benchmark --dataset <agentrx|codetracebench> --labels <path> --trace-dir <path> [--artifact-dir <path>] --out <dir> ...\n        Compare an empty baseline with a real-model trace analyst on public labels.\n        Run with --help for all required inputs and controls.\n  supervisor-run report <runDir> [--format headline|markdown|json]\n        Report one Runtime or loops supervisor run directory. The status comes\n        from Runtime's own result.json / failure.json; every missing measurement\n        stays named. Exits 1 when the directory cannot be read.\n  version\n        Print server + wire-protocol version JSON.\n\nJudge provider:\n  Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n  OPENAI_* and TANGLE_* equivalents are also accepted. This binary is the only\n  place that reads a provider credential; the library never does. Without both\n  a base URL and a key, /v1/judge refuses with llm_not_configured.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n  const { command, positional, flags } = parseArgs(process.argv.slice(2))\n  assertKnownFlags(command, flags)\n\n  if (command === 'analyst-benchmark' && flags.help === 'true') {\n    return await runAnalystBenchmarkCommand(process.argv.slice(3))\n  }\n  if (flags.help === 'true') {\n    process.stdout.write(`${HELP}\\n`)\n    return 0\n  }\n\n  switch (command) {\n    case 'serve': {\n      const port = parsePort(flags.port ?? '5005')\n      const host = flags.host ?? '127.0.0.1'\n      const llm = resolveCliLlmConfig()\n      const { server } = await startServerAsync({\n        port,\n        host,\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n      // Keep process alive on SIGINT/SIGTERM\n      const shutdown = (sig: string) => {\n        // eslint-disable-next-line no-console\n        console.log(`[agent-eval] received ${sig}, shutting down`)\n        server.close(() => process.exit(0))\n        // Force exit after 5s if close hangs\n        setTimeout(() => process.exit(1), 5000).unref()\n      }\n      process.on('SIGINT', () => shutdown('SIGINT'))\n      process.on('SIGTERM', () => shutdown('SIGTERM'))\n      // Block forever\n      await new Promise(() => {})\n      return 0\n    }\n    case 'rpc': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcOnce(method, {\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n    }\n    case 'rpc-batch': {\n      const [method] = positional\n      const llm = resolveCliLlmConfig()\n      return await runRpcBatch(method, {\n        ...(llm.chat ? { chat: llm.chat } : {}),\n        ...(llm.model ? { judgeModel: llm.model } : {}),\n      })\n    }\n    case 'openapi': {\n      const out = flags.out ?? 'openapi.json'\n      const spec = buildOpenApi(handleVersion().version)\n      writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n      // eslint-disable-next-line no-console\n      console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n      return 0\n    }\n    case 'rollout-release': {\n      // The subcommand owns its own flag grammar (multi-value --formats,\n      // boolean --include-proposers) — pass raw argv through untouched.\n      return await runRolloutReleaseCli(process.argv.slice(3))\n    }\n    case 'analyst-benchmark': {\n      return await runAnalystBenchmarkCommand(process.argv.slice(3))\n    }\n    case 'supervisor-run': {\n      // The subcommand owns its own flag grammar; pass raw argv through.\n      return await runSupervisorRunCommand(process.argv.slice(3))\n    }\n    case 'version': {\n      process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n      return 0\n    }\n    case '--version': {\n      process.stdout.write(`${handleVersion().version}\\n`)\n      return 0\n    }\n    case 'help':\n    case '--help':\n    case '-h':\n    case '':\n      process.stdout.write(`${HELP}\\n`)\n      return 0\n    default:\n      process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n      return 1\n  }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n  serve: new Set(['help', 'host', 'port']),\n  rpc: new Set(['help']),\n  'rpc-batch': new Set(['help']),\n  openapi: new Set(['help', 'out']),\n  version: new Set(['help']),\n  help: new Set(),\n  '--help': new Set(),\n  '-h': new Set(),\n  '--version': new Set(),\n  '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n  // These subcommands parse their own argv.\n  if (\n    command === 'rollout-release' ||\n    command === 'analyst-benchmark' ||\n    command === 'supervisor-run'\n  ) {\n    return\n  }\n  const allowed = FLAGS_BY_COMMAND[command]\n  if (!allowed) return\n  const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n  if (unknown.length > 0) {\n    throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n  }\n}\n\nfunction parsePort(raw: string): number {\n  const port = Number(raw)\n  if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n    throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n  }\n  return port\n}\n\nmain()\n  .then((code) => process.exit(code))\n  .catch((err) => {\n    // eslint-disable-next-line no-console\n    console.error('[agent-eval] cli error:', err)\n    process.exit(1)\n  })\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;AAiCA,SAAgB,wBACd,MAAyB,QAAQ,KACH;CAC9B,MAAM,kBAAkB,SAAS,IAAI,uBAAuB;CAC5D,MAAM,iBAAiB,SAAS,IAAI,sBAAsB;CAC1D,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,KAAA,OAC7C,eAAe,mCAAmC,KAAA;CACrD,MAAM,SAAS,kBAAkB,gBAAgB;CACjD,MAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;CAE/F,IAAI,CAAC,WAAW,CAAC,QAAQ,OAAO,KAAA;CAChC,OAAO;EAAE;EAAS;EAAQ,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;CAAG;AACxD;AAEA,SAAgB,oBAAoB,MAAyB,QAAQ,KAAmB;CACtF,MAAM,QAAQ,wBAAwB,GAAG;CACzC,IAAI,CAAC,OAAO,OAAO,CAAC;CACpB,OAAO;EACL,MAAM,cAAc;GAAE,SAAS,MAAM;GAAS,QAAQ,MAAM;EAAO,GAAG,MAAM,KAAK;EACjF,GAAI,MAAM,QAAQ,EAAE,OAAO,MAAM,MAAM,IAAI,CAAC;CAC9C;AACF;;;;;;;;;AAUA,SAAS,cACP,OACA,cACY;CACZ,MAAM,SAAS,iBAAiB;EAC9B,WAAW;EACX,SAAS,MAAM;EACf,QAAQ,MAAM;EACd,GAAI,eAAe,EAAE,aAAa,IAAI,CAAC;CACzC,CAAC;CACD,OAAO;EACL,GAAG;EACH,OAAO,KAAK,aAAa;GACvB,IAAI,CAAC,IAAI,SAAS,CAAC,cACjB,MAAM,IAAI,MACR,4EACF;GAEF,OAAO,OAAO,KAAK,KAAK,QAAQ;EAClC;CACF;AACF;AAEA,SAAS,SAAS,OAA+C;CAC/D,MAAM,UAAU,OAAO,KAAK;CAC5B,OAAO,UAAU,UAAU,KAAA;AAC7B;;;;;;;;;;;;;;;;;AClEA,SAAS,UAAU,MAAsB;CACvC,MAAM,CAAC,SAAS,GAAG,QAAQ;CAC3B,MAAM,aAAuB,CAAC;CAC9B,MAAM,QAAgC,CAAC;CACvC,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,IAAI,IAAI,WAAW,IAAI,GAAG;GACxB,MAAM,MAAM,IAAI,MAAM,CAAC;GACvB,MAAM,WAAW,IAAI,QAAQ,GAAG;GAChC,IAAI,YAAY,GAAG;IACjB,MAAM,IAAI,MAAM,GAAG,QAAQ,KAAK,IAAI,MAAM,WAAW,CAAC;IACtD;GACF;GACA,MAAM,MAAM;GACZ,IAAI,QAAQ,QAAQ;IAClB,MAAM,OAAO;IACb;GACF;GACA,MAAM,OAAO,KAAK,IAAI;GACtB,IAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;IAC1C,MAAM,OAAO;IACb;GACF,OACE,MAAM,OAAO;EAEjB,OAAO,IAAI,QAAQ,MACjB,MAAM,OAAO;OAEb,WAAW,KAAK,GAAG;CAEvB;CACA,OAAO;EAAE,SAAS,WAAW;EAAQ;EAAY;CAAM;AACzD;AAEA,MAAM,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiCb,eAAe,OAAwB;CACrC,MAAM,EAAE,SAAS,YAAY,UAAU,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;CACtE,iBAAiB,SAAS,KAAK;CAE/B,IAAI,YAAY,uBAAuB,MAAM,SAAS,QACpD,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;CAE/D,IAAI,MAAM,SAAS,QAAQ;EACzB,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;EAChC,OAAO;CACT;CAEA,QAAQ,SAAR;EACE,KAAK,SAAS;GACZ,MAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;GAC3C,MAAM,OAAO,MAAM,QAAQ;GAC3B,MAAM,MAAM,oBAAoB;GAChC,MAAM,EAAE,WAAW,MAAM,iBAAiB;IACxC;IACA;IACA,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;GAED,MAAM,YAAY,QAAgB;IAEhC,QAAQ,IAAI,yBAAyB,IAAI,gBAAgB;IACzD,OAAO,YAAY,QAAQ,KAAK,CAAC,CAAC;IAElC,iBAAiB,QAAQ,KAAK,CAAC,GAAG,GAAI,CAAC,CAAC,MAAM;GAChD;GACA,QAAQ,GAAG,gBAAgB,SAAS,QAAQ,CAAC;GAC7C,QAAQ,GAAG,iBAAiB,SAAS,SAAS,CAAC;GAE/C,MAAM,IAAI,cAAc,CAAC,CAAC;GAC1B,OAAO;EACT;EACA,KAAK,OAAO;GACV,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,WAAW,QAAQ;IAC9B,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;EACH;EACA,KAAK,aAAa;GAChB,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,YAAY,QAAQ;IAC/B,GAAI,IAAI,OAAO,EAAE,MAAM,IAAI,KAAK,IAAI,CAAC;IACrC,GAAI,IAAI,QAAQ,EAAE,YAAY,IAAI,MAAM,IAAI,CAAC;GAC/C,CAAC;EACH;EACA,KAAK,WAAW;GACd,MAAM,MAAM,MAAM,OAAO;GACzB,MAAM,OAAO,aAAa,cAAc,CAAC,CAAC,OAAO;GACjD,cAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,EAAE,KAAK,OAAO;GAEhE,QAAQ,IAAI,0CAA0C,KAAK;GAC3D,OAAO;EACT;EACA,KAAK,mBAGH,OAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAEzD,KAAK,qBACH,OAAO,MAAM,2BAA2B,QAAQ,KAAK,MAAM,CAAC,CAAC;EAE/D,KAAK,kBAEH,OAAO,MAAM,wBAAwB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAE5D,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,EAAE,GAAG;GACpE,OAAO;EAET,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,cAAc,CAAC,CAAC,QAAQ,GAAG;GACnD,OAAO;EAET,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;GAChC,OAAO;EACT;GACE,QAAQ,OAAO,MAAM,oBAAoB,QAAQ,IAAI,KAAK,GAAG;GAC7D,OAAO;CACX;AACF;AAEA,MAAM,mBAAwD;CAC5D,uBAAO,IAAI,IAAI;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,qBAAK,IAAI,IAAI,CAAC,MAAM,CAAC;CACrB,6BAAa,IAAI,IAAI,CAAC,MAAM,CAAC;CAC7B,yBAAS,IAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;CAChC,yBAAS,IAAI,IAAI,CAAC,MAAM,CAAC;CACzB,sBAAM,IAAI,IAAI;CACd,0BAAU,IAAI,IAAI;CAClB,sBAAM,IAAI,IAAI;CACd,6BAAa,IAAI,IAAI;CACrB,oBAAI,IAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;CAE9E,IACE,YAAY,qBACZ,YAAY,uBACZ,YAAY,kBAEZ;CAEF,MAAM,UAAU,iBAAiB;CACjC,IAAI,CAAC,SAAS;CACd,MAAM,UAAU,OAAO,KAAK,KAAK,CAAC,CAAC,QAAQ,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;CACtE,IAAI,QAAQ,SAAS,GACnB,MAAM,IAAI,MAAM,oBAAoB,WAAW,OAAO,MAAM,QAAQ,IAAI;AAE5E;AAEA,SAAS,UAAU,KAAqB;CACtC,MAAM,OAAO,OAAO,GAAG;CACvB,IAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAChD,MAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,GAAG;CAE9F,OAAO;AACT;AAEA,KAAK,CAAC,CACH,MAAM,SAAS,QAAQ,KAAK,IAAI,CAAC,CAAC,CAClC,OAAO,QAAQ;CAEd,QAAQ,MAAM,2BAA2B,GAAG;CAC5C,QAAQ,KAAK,CAAC;AAChB,CAAC"}

package/dist/contract/index.js

@@ -11,7 +11,7 @@ import { C as inMemoryCampaignStorage, S as fsCampaignStorage } from "../externa
 import { a as heldoutSignificance, s as decidePairedPromotion } from "../power-preflight-CFXm0Vjo.js";
 import { i as makeProposalFinding } from "../types-DQ0e2E7y.js";
 import { G as classifyOtlpSpanRole, K as isOtlpModelCall } from "../kind-factory-BLvL-E44.js";
-import { t as buildDefaultAnalystRegistry } from "../default-registry-CrAp0pYq.js";
+import { t as buildDefaultAnalystRegistry } from "../default-registry-DBqVI4pq.js";
 import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js";
 import { t as extractUsage } from "../extract-usage-BrQ8mCLX.js";
 import { t as createChatClient } from "../chat-client-CkmjYlfB.js";

package/dist/{default-registry-CrAp0pYq.js → default-registry-DBqVI4pq.js}

@@ -4,7 +4,7 @@ import { a as assertValidAnalystUsageReceipt, c as validateUsageSettlementTimeou
 import { J as snapshotExactExecutionComponentIdentity, O as spanEpochMillis, Y as snapshotExactExecutionPlan, t as createTraceAnalyst, z as findingSubjectGrammarPromptFor } from "./kind-factory-BLvL-E44.js";
 import { LLM_CONTEXT_TOKENS, LLM_INPUT_TOKEN_ATTR_KEYS, LLM_OUTPUT_TOKEN_ATTR_KEYS, TOOL_NAME_ATTR_KEYS } from "./trace-attributes.js";
 import { t as executionTrackByLane } from "./execution-tracks-CpgFPpS5.js";
-import { t as analyzeSupervisorRunIntegrity } from "./integrity-BWywb34E.js";
+import { t as analyzeSupervisorRunIntegrity } from "./integrity-DsHWCebQ.js";
 import { randomUUID } from "node:crypto";
 import { z } from "zod";
 //#region src/feedback-trajectory-review.ts
@@ -2363,4 +2363,4 @@ function buildDefaultAnalystRegistry(options = {}) {
 //#endregion
 export { completedAnalystReviewQuality as _, KNOWLEDGE_POISONING_KIND_SPEC as a, validateAnalystReviewDecisions as b, FAILURE_MODE_KIND_SPEC as c, emitControlIntegrityFindings as d, behavioralAnalyst as f, assertUniqueFindingIds as g, analystRunDigest as h, DEFAULT_TRACE_ANALYST_KINDS as i, CONTROL_INTEGRITY_ANALYST as l, analystFindingDigest as m, AnalystRegistry as n, KNOWLEDGE_GAP_KIND_SPEC as o, deriveEfficiencyFindings as p, ExactAnalystRunExecutionError as r, IMPROVEMENT_KIND_SPEC as s, buildDefaultAnalystRegistry as t, ControlIntegrityAnalyst as u, readAnalystReview as v, snapshotAnalystRun as y };
 
-//# sourceMappingURL=default-registry-CrAp0pYq.js.map
+//# sourceMappingURL=default-registry-DBqVI4pq.js.map