perchai-cli 2.4.14 → 2.4.16

package/dist/perch.mjs

@@ -83131,7 +83131,6 @@ function truncateHistoryLine(value, max2) {
 }
 var init_operatorTruth = __esm({
   "features/perchTerminal/runtime/operatorTruth.ts"() {
-    "use strict";
   }
 });
 
@@ -134243,7 +134242,7 @@ function validateArgs(name, args) {
           return "run_sandbox_code.command must be a non-empty string when provided.";
         if (args.language !== void 0 && args.language !== "python" && args.language !== "node")
           return 'run_sandbox_code.language must be either "python" or "node" when provided.';
-        if (args.code !== void 0 && !hasCode)
+        if (args.code !== void 0 && !hasCode && !hasCommand)
           return "run_sandbox_code.code must be a non-empty string when provided.";
       }
       if (args.label !== void 0 && typeof args.label !== "string")
@@ -199728,7 +199727,6 @@ function containsBrowserDeliveryTask(tasks) {
 var BROWSER_DELIVERY_ROLE_IDS;
 var init_browserDeliveryLock = __esm({
   "features/perchTerminal/agentPlatform/browserDeliveryLock.ts"() {
-    "use strict";
     BROWSER_DELIVERY_ROLE_IDS = /* @__PURE__ */ new Set([
       "doc_writer",
       "email_sender",
@@ -213392,11 +213390,14 @@ var init_config2 = __esm({
 
 // features/perchTerminal/knowledge/knowledgeApiClient.ts
 async function searchKnowledgeViaApi(input, options = {}) {
-  const route = options.route ?? "/api/perch-terminal/knowledge/search";
+  const route = resolveKnowledgeRoute(options.route ?? "/api/perch-terminal/knowledge/search");
   const fetchImpl = options.fetchImpl ?? fetch;
+  const headers = { "content-type": "application/json" };
+  const token = process.env.PERCH_MODEL_CALL_PROXY_TOKEN?.trim();
+  if (token) headers.authorization = `Bearer ${token}`;
   const response = await fetchImpl(route, {
     method: "POST",
-    headers: { "content-type": "application/json" },
+    headers,
     body: JSON.stringify(input)
   });
   const payload = await response.json().catch(() => ({}));
@@ -213408,6 +213409,13 @@ async function searchKnowledgeViaApi(input, options = {}) {
   }
   return payload;
 }
+function resolveKnowledgeRoute(route) {
+  if (!route.startsWith("/")) return route;
+  if (typeof window !== "undefined") return route;
+  const appUrl = process.env.PERCH_MODEL_CALL_PROXY_URL?.trim() || process.env.PERCH_CLI_APP_URL?.trim() || process.env.PERCH_APP_URL?.trim();
+  if (!appUrl) return route;
+  return new URL(route, appUrl).toString();
+}
 var KnowledgeApiClientError;
 var init_knowledgeApiClient = __esm({
   "features/perchTerminal/knowledge/knowledgeApiClient.ts"() {
@@ -214630,13 +214638,22 @@ async function computeLiveSignals(store, strategyId, asOf) {
   }
   return out;
 }
+function ledgerFields(result2) {
+  return result2.error ? { ledgerRecorded: result2.recorded, ledgerError: result2.error } : { ledgerRecorded: result2.recorded };
+}
 async function recordLedger(ctx, row) {
-  if (!ctx.supabase) return false;
+  if (!ctx.supabase) {
+    return { recorded: false, error: "supabase_unavailable" };
+  }
   try {
     const { error } = await ctx.supabase.from("perch_ai_market_backtest_runs").insert(row);
-    return !error;
-  } catch {
-    return false;
+    if (error) return { recorded: false, error: error.message };
+    return { recorded: true };
+  } catch (error) {
+    return {
+      recorded: false,
+      error: error instanceof Error ? error.message : String(error)
+    };
   }
 }
 var NOT_ADVICE, getMarketSignalTool, queryMarketSignalLogTool, explainMarketSignalTool, listMarketStrategiesTool, MAX_PERIOD_DAYS, ISO_DATE, runMarketBacktestTool, getMarketTrackRecordTool, marketDeskTools;
@@ -214900,7 +214917,8 @@ var init_marketDesk = __esm({
           holdout_acknowledged: acknowledgeHoldout,
           initiator: "tool",
           workspace_id: ctx.workspaceId ?? null,
-          thread_id: ctx.threadId ?? null
+          thread_id: ctx.threadId ?? null,
+          user_id: ctx.userId ?? null
         };
         const wf = args.walkForward && typeof args.walkForward === "object" ? args.walkForward : void 0;
         try {
@@ -214923,10 +214941,10 @@ var init_marketDesk = __esm({
               paramGrid
             });
             if (!result3.ok) {
-              await recordLedger(ctx, { ...ledgerBase, status: "failed", error: result3.error, holdout_overlap: false });
-              return { ok: false, errorCode: "walk_forward_failed", message: result3.error };
+              const ledger3 = await recordLedger(ctx, { ...ledgerBase, status: "failed", error: result3.error, holdout_overlap: false });
+              return { ok: false, errorCode: "walk_forward_failed", message: result3.error, ...ledgerFields(ledger3) };
             }
-            const ledgerRecorded2 = await recordLedger(ctx, {
+            const ledger2 = await recordLedger(ctx, {
               ...ledgerBase,
               status: "completed",
               holdout_overlap: to >= resolveHoldoutStart(),
@@ -214951,17 +214969,17 @@ var init_marketDesk = __esm({
                 testMaxDrawdown: round(f.testMetrics.maxDrawdown)
               })),
               oosMetrics: compactMetrics(result3.oosMetrics),
-              ledgerRecorded: ledgerRecorded2,
+              ...ledgerFields(ledger2),
               note: "Out-of-sample (oosMetrics) is the only number that counts; train-window performance is selection, not evidence.",
               disclaimer: NOT_ADVICE
             };
           }
           const result2 = await runBacktest(store, backtestCfg);
           if (!result2.ok) {
-            await recordLedger(ctx, { ...ledgerBase, status: "failed", error: result2.error, holdout_overlap: false });
-            return { ok: false, errorCode: "backtest_failed", message: result2.error };
+            const ledger2 = await recordLedger(ctx, { ...ledgerBase, status: "failed", error: result2.error, holdout_overlap: false });
+            return { ok: false, errorCode: "backtest_failed", message: result2.error, ...ledgerFields(ledger2) };
           }
-          const ledgerRecorded = await recordLedger(ctx, {
+          const ledger = await recordLedger(ctx, {
             ...ledgerBase,
             status: "completed",
             holdout_overlap: result2.holdout.overlapped,
@@ -214990,13 +215008,13 @@ var init_marketDesk = __esm({
             signalCounts: result2.signalCounts,
             equityCurve: downsample(result2.equityCurve, 60).map((p) => ({ date: p.date, equity: round(p.equity, 4) })),
             trades: { closed: result2.trades.filter((t) => t.netReturn !== null).length, open: result2.trades.filter((t) => t.exitDate === null && t.netReturn === null).length },
-            ledgerRecorded,
+            ...ledgerFields(ledger),
             note: "Hypothetical backtest with modeled fees/slippage; past simulated performance does not predict future results. Every run (including this one) is recorded in the experiment ledger.",
             disclaimer: NOT_ADVICE
           };
         } catch (error) {
           if (error instanceof HoldoutViolationError) {
-            await recordLedger(ctx, {
+            const ledger2 = await recordLedger(ctx, {
               ...ledgerBase,
               status: "failed",
               error: "holdout_violation",
@@ -215007,12 +215025,13 @@ var init_marketDesk = __esm({
               errorCode: "holdout_violation",
               message: error.message,
               holdoutStart: error.holdoutStart,
-              hint: "The holdout window is reserved for final evaluation. Re-run with acknowledgeHoldout: true ONLY if the user explicitly wants to spend a holdout look; the run gets flagged in the ledger either way."
+              hint: "The holdout window is reserved for final evaluation. Re-run with acknowledgeHoldout: true ONLY if the user explicitly wants to spend a holdout look; the run gets flagged in the ledger either way.",
+              ...ledgerFields(ledger2)
             };
           }
           const message = error instanceof Error ? error.message : String(error);
-          await recordLedger(ctx, { ...ledgerBase, status: "failed", error: message, holdout_overlap: false });
-          return { ok: false, errorCode: "backtest_error", message };
+          const ledger = await recordLedger(ctx, { ...ledgerBase, status: "failed", error: message, holdout_overlap: false });
+          return { ok: false, errorCode: "backtest_error", message, ...ledgerFields(ledger) };
         }
       }
     };
@@ -215190,11 +215209,22 @@ var init_registry5 = __esm({
 async function executeRegisteredTool(name, args, ctx) {
   const mod = getRegisteredTool(name);
   if (!mod) return { handled: false };
+  if (MARKET_DESK_TOOL_NAMES.has(name) && !isMarketDeskEnabled()) {
+    return {
+      handled: true,
+      result: {
+        ok: false,
+        errorCode: "market_desk_disabled",
+        message: "Market Desk tools are disabled. Set PERCH_MARKET_DESK=1 to enable them."
+      }
+    };
+  }
   return { handled: true, result: await mod.handler(args, ctx) };
 }
 var init_executeTool = __esm({
   "features/perchTerminal/runtime/toolSystem/executeTool.ts"() {
     "use strict";
+    init_marketDeskAccess();
     init_registry5();
   }
 });
@@ -219674,17 +219704,18 @@ async function runLiveAgentsLoop(input) {
     }
     const mode = effectiveChatMode;
     const loopTools = tools;
-    const loopSystemPrompt = appendLiveTurnContract(context.systemPrompt, {
+    const sandboxExecutionRequired = sandboxExecutionRequiredByUserText(turn.trimmedInput, loopTools);
+    const loopSystemPromptBase = appendLiveTurnContract(context.systemPrompt, {
       chatMode: mode,
       toolsAvailable: loopTools.length > 0,
       userObjective: turn.trimmedInput,
       planGateRejected
     });
+    const loopSystemPrompt = sandboxExecutionRequired ? appendSandboxExecutionRequiredContract(loopSystemPromptBase) : loopSystemPromptBase;
     const threadSession = turn.threadId ? getThreadSessionFromMemory(turn.threadId) : null;
     const contextCompaction = threadSession?.contextCompaction ?? null;
     const approvedToolCall = approvedToolCallFromState(workflowState);
     const runLoop = input.loopRunner ?? runModelToolLoop;
-    const sandboxExecutionRequired = sandboxExecutionRequiredByUserText(turn.trimmedInput, loopTools);
     const loopInput = {
       signal: deps.signal ?? void 0,
       lane: loopLane,
@@ -219715,7 +219746,7 @@ async function runLiveAgentsLoop(input) {
       forcedInitialToolCall: approvedToolCall ?? turn.forcedInitialToolCall ?? null,
       forcedInitialToolCallApproved: Boolean(approvedToolCall),
       stopAfterForcedInitialToolCall: Boolean(approvedToolCall),
-      forceToolUse: turn.forceToolUse === true || sandboxExecutionRequired,
+      forceToolUse: turn.forceToolUse === true,
       mcpTools: turn.mcpTools ?? [],
       maxIterations: turnHasBrowserOperatorTools(loopTools) ? 24 : turn.coordinatorSessionActive ? 20 : void 0,
       attachOperatorScreenshots: turnHasBrowserOperatorTools(loopTools),
@@ -219812,10 +219843,11 @@ async function runLiveAgentsLoop(input) {
     if (sandboxExecutionRequired && loopResult.ok && !hasRunSandboxCodeToolCall(loopResult)) {
       loopResult = await runLoop({
         ...loopInput,
-        forceToolUse: true,
-        systemPrompt: `${loopInput.systemPrompt.trim()}
-
-Sandbox execution is explicitly required by the user. This turn is incomplete until you call run_sandbox_code. Use the command field when possible; pass source file paths in sources and read copied files from input/. The sandbox tool result already captures stdout/stderr and produced files; do not try to read output/report.json from the host filesystem after the run. Do not mention internal model, provider, or lane details in the user-facing answer. Do not finish with prose before a sandbox run.`
+        forceToolUse: false,
+        systemPrompt: appendSandboxExecutionRequiredContract(
+          loopInput.systemPrompt,
+          true
+        )
       });
     }
     if (sandboxExecutionRequired && loopResult.ok && !hasRunSandboxCodeToolCall(loopResult)) {
@@ -219946,6 +219978,12 @@ function sandboxExecutionRequiredByUserText(userText, tools) {
   if (!mentionsSandbox) return false;
   return /\b(use|run|execute|call|primary|required|mandatory|must|final numbers|must come from|do not stop)\b/.test(normalized) || /sandbox.{0,80}(primary|required|mandatory|must|execute|run|numbers|code)/.test(normalized) || /(primary|required|mandatory|must|execute|run|numbers|code).{0,80}sandbox/.test(normalized);
 }
+function appendSandboxExecutionRequiredContract(systemPrompt, retry = false) {
+  const prefix = retry ? "The previous attempt did not call run_sandbox_code." : "Sandbox execution is explicitly required by the user.";
+  return `${systemPrompt.trim()}
+
+${prefix} This turn is incomplete until you call run_sandbox_code. Use the command field when possible; pass source file paths in sources and read copied files from input/. The sandbox tool result already captures stdout/stderr and produced files; do not try to read output/report.json from the host filesystem after the run. Do not mention internal model, provider, or lane details in the user-facing answer. Do not finish with prose before a sandbox run.`;
+}
 function hasRunSandboxCodeToolCall(loopResult) {
   return loopResult.toolCalls.some((call) => call.toolName === TOOL_NAMES.runSandboxCode);
 }
@@ -222782,7 +222820,9 @@ function maybeWarnCategoryDrift(input) {
   const drift = sum - input.headlineTokens;
   const allowed = Math.max(256, Math.ceil(input.wireOverheadTokens ?? 0) + 64);
   if (Math.abs(drift) <= allowed) return;
-  if (typeof process !== "undefined" && process.env.NODE_ENV === "production") return;
+  if (typeof process === "undefined" || process.env.PERCH_CONTEXT_METER_DEBUG !== "1") {
+    return;
+  }
   console.warn(
     `[context-meter] Category total drift in ${input.source}: categories=${sum}, headline=${input.headlineTokens}, drift=${drift}`
   );

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "perchai-cli",
-  "version": "2.4.14",
+  "version": "2.4.16",
   "description": "Perch AI command-line interface",
   "bin": {
     "perch": "bin/perch"
@@ -12,7 +12,9 @@
     "LICENSE"
   ],
   "dependencies": {
-    "@napi-rs/canvas": "^0.1.100"
+    "@mozilla/readability": "^0.6.0",
+    "@napi-rs/canvas": "^0.1.100",
+    "jsdom": "^29.1.1"
   },
   "engines": {
     "node": ">=20"