@mrciphersmith/keryx 0.3.17 → 0.3.18

package/dist/cli.js

@@ -17641,7 +17641,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@mrciphersmith/keryx",
-    version: "0.3.17",
+    version: "0.3.18",
     description: "Version-controlled project context for AI coding agents: code graph, architecture wiki, project memory, relevant tests, quality signals, and task flows.",
     private: false,
     publishConfig: {
@@ -28955,7 +28955,7 @@ function retryableFor2(kind, fallback) {
   const concrete = defaultRetryable(kind);
   return concrete === undefined ? fallback : concrete;
 }
-function mergeUsage2(promptTokens, completionTokens, totalTokens) {
+function mergeUsage2(promptTokens, completionTokens, totalTokens, cacheReadTokens) {
   const usage = { exact: true };
   if (promptTokens !== undefined) {
     usage.inputTokens = promptTokens;
@@ -28968,6 +28968,9 @@ function mergeUsage2(promptTokens, completionTokens, totalTokens) {
   } else if (promptTokens !== undefined || completionTokens !== undefined) {
     usage.totalTokens = (promptTokens ?? 0) + (completionTokens ?? 0);
   }
+  if (cacheReadTokens !== undefined) {
+    usage.cacheReadTokens = cacheReadTokens;
+  }
   return usage;
 }
 function classifyHttpError2(status, headers, code) {
@@ -29394,6 +29397,14 @@ ${message.content}` });
         }
         return "idx:0";
       };
+      const pendingToolCallStillAccumulating = () => {
+        for (const acc of pendingTools.values()) {
+          if (!acc.ended) {
+            return acc.id;
+          }
+        }
+        return;
+      };
       const flushPendingToolEnds = () => {
         for (const acc of pendingTools.values()) {
           if (acc.ended) {
@@ -29487,7 +29498,7 @@ ${message.content}` });
               const usage = asRecord4(data.usage);
               bodies.push({
                 kind: "usage_update",
-                usage: mergeUsage2(asNumber3(usage.prompt_tokens), asNumber3(usage.completion_tokens), asNumber3(usage.total_tokens))
+                usage: mergeUsage2(asNumber3(usage.prompt_tokens), asNumber3(usage.completion_tokens), asNumber3(usage.total_tokens), asNumber3(asRecord4(usage.prompt_tokens_details).cached_tokens))
               });
             }
             const choice0 = asRecord4(asArray(data.choices)[0]);
@@ -29600,6 +29611,7 @@ ${message.content}` });
           parser.push(trailing);
         }
         const torn = parser.flush();
+        const pendingToolCallId = pendingToolCallStillAccumulating();
         if (torn.length > 0) {
           malformed = {
             kind: "malformed",
@@ -29612,8 +29624,14 @@ ${message.content}` });
             retryable: retryableFor2("malformed", false),
             message: "empty response body"
           };
+        } else if (pendingToolCallId !== undefined) {
+          malformed = {
+            kind: "malformed",
+            retryable: retryableFor2("malformed", false),
+            message: `${this.label} SSE stream ended while a tool call was still accumulating`,
+            detail: { pendingToolCallId }
+          };
         } else {
-          flushPendingToolEnds();
           if (sawStart && sawFinish) {
             flushThinkTagParser();
             flushFieldTagStripper();
@@ -30272,6 +30290,16 @@ class GeminiProvider {
             break;
           }
           const chunk = asRecord5(parsed);
+          const errorEnvelope = asRecord5(chunk.error);
+          if (Object.keys(errorEnvelope).length > 0) {
+            const classified = classifyGeminiError(asNumber4(errorEnvelope.code) ?? 0, asString4(errorEnvelope.status));
+            const envelopeMessage = asString4(errorEnvelope.message);
+            malformed = {
+              ...classified,
+              message: redact(envelopeMessage !== undefined && envelopeMessage.length > 0 ? envelopeMessage : "Gemini stream carried an in-band error envelope")
+            };
+            break;
+          }
           if (!sawFirstChunk) {
             sawFirstChunk = true;
             bodies.push({ kind: "model_start" });
@@ -30295,7 +30323,7 @@ class GeminiProvider {
                 bodies.push({ kind: "text_delta", text });
               }
             } else if (callName !== undefined) {
-              const callId = asString4(functionCall.id) ?? callName;
+              const callId = asString4(functionCall.id) ?? `${callName}#${functionCallIndexInRound}`;
               const argsInput = JSON.stringify(asRecord5(functionCall.args));
               bodies.push({ kind: "tool_call_start", toolCallId: callId, toolName: callName });
               bodies.push({ kind: "tool_call_end", toolCallId: callId, input: argsInput });
@@ -30512,7 +30540,7 @@ function retryableFor4(kind, fallback) {
   const concrete = defaultRetryable(kind);
   return concrete === undefined ? fallback : concrete;
 }
-function mergeUsage4(inputTokens, outputTokens, totalTokens) {
+function mergeUsage4(inputTokens, outputTokens, totalTokens, cacheReadTokens) {
   const usage = { exact: true };
   if (inputTokens !== undefined) {
     usage.inputTokens = inputTokens;
@@ -30525,6 +30553,9 @@ function mergeUsage4(inputTokens, outputTokens, totalTokens) {
   } else if (inputTokens !== undefined || outputTokens !== undefined) {
     usage.totalTokens = (inputTokens ?? 0) + (outputTokens ?? 0);
   }
+  if (cacheReadTokens !== undefined) {
+    usage.cacheReadTokens = cacheReadTokens;
+  }
   return usage;
 }
 function extractErrorFields(data) {
@@ -30942,9 +30973,10 @@ class OpenAiProvider {
               const outputTokens = asNumber5(usage.output_tokens);
               const totalTokens = asNumber5(usage.total_tokens);
               const reasoningTokens = asNumber5(asRecord6(usage.output_tokens_details).reasoning_tokens);
+              const cacheReadTokens = asNumber5(asRecord6(usage.input_tokens_details).cached_tokens);
               const usageBody = {
                 kind: "usage_update",
-                usage: mergeUsage4(inputTokens, outputTokens, totalTokens)
+                usage: mergeUsage4(inputTokens, outputTokens, totalTokens, cacheReadTokens)
               };
               if (reasoningTokens !== undefined) {
                 usageBody.unknownExtensions = { "openai.reasoning_tokens": reasoningTokens };
@@ -58148,8 +58180,10 @@ async function runAgentTurn(io, deps, history, userLine, options = {}) {
 
 ${buildControlMarkerInstruction(controlNonce)}`
   };
+  const caughtToolErrors = [];
   try {
-    const result = await runAgentTurnCore(io, turnDeps, history, userLine, options);
+    const coreResult = await runAgentTurnCore(io, turnDeps, history, userLine, options, caughtToolErrors);
+    const result = caughtToolErrors.length > 0 ? { ...coreResult, caughtToolErrors } : coreResult;
     await fireStopHookBestEffort(io, deps, result);
     return result;
   } finally {
@@ -58231,7 +58265,7 @@ function computeReasoningDurationMs(startedAt, endedAt) {
   const endMs = Date.parse(endedAt);
   return !Number.isNaN(startMs) && !Number.isNaN(endMs) && endMs >= startMs ? endMs - startMs : undefined;
 }
-async function runAgentTurnCore(io, deps, history, userLine, options = {}) {
+async function runAgentTurnCore(io, deps, history, userLine, options = {}, caughtToolErrors = []) {
   const now = deps.now ?? (() => new Date().toISOString());
   const controlNonce = deps.controlNonce ?? generateControlNonce();
   const scrub = (text) => scrubControlNonce(text, controlNonce);
@@ -58821,7 +58855,7 @@ ${shown.join(`
     }
     const spawnConcurrencyCandidates = calls.filter((call) => call.name === "spawn_subagent" && reservationByCallId.get(call.id)?.ok === true);
     const untrustedGateBlocksSpawns = untrustedContentSeen || batchContainsUntrustedWeb;
-    const concurrentSpawnResults = spawnConcurrencyCandidates.length >= 2 && !untrustedGateBlocksSpawns && maxToolCalls === undefined ? await runConcurrentSpawnBatch(spawnConcurrencyCandidates, toolByName, io, deps, hasInvocationCapacity, reserveInvocation, signal) : undefined;
+    const concurrentSpawnResults = spawnConcurrencyCandidates.length >= 2 && !untrustedGateBlocksSpawns && maxToolCalls === undefined ? await runConcurrentSpawnBatch(spawnConcurrencyCandidates, toolByName, io, deps, hasInvocationCapacity, reserveInvocation, signal, caughtToolErrors) : undefined;
     const answeredToolIds = new Set;
     let anchorsToAnnounce;
     let repeatedFailureHint;
@@ -58921,7 +58955,19 @@ ${shown.join(`
       if (toolByName.has(call.name)) {
         turnExecutedToolCall = true;
       }
-      const result = concurrentSpawnResults?.get(call.id) ?? await executeCall(call, toolByName, io.requestApproval, io.permissionMode, io.readOnly, io.onAutoApproved, hasInvocationCapacity, reserveInvocation, invocationBudget.maxCalls, signal, deps.busLeases, deps.hardDeny === undefined ? undefined : { check: deps.hardDeny, onDenied: io.onUnattendedDenial }, deps.hooks);
+      const precomputedResult = concurrentSpawnResults?.get(call.id);
+      let result;
+      if (precomputedResult !== undefined) {
+        result = precomputedResult;
+      } else {
+        try {
+          result = await executeCall(call, toolByName, io.requestApproval, io.permissionMode, io.readOnly, io.onAutoApproved, hasInvocationCapacity, reserveInvocation, invocationBudget.maxCalls, signal, deps.busLeases, deps.hardDeny === undefined ? undefined : { check: deps.hardDeny, onDenied: io.onUnattendedDenial }, deps.hooks);
+        } catch (err) {
+          const message = err instanceof Error ? err.message : String(err);
+          result = { output: `${call.name} failed: ${message}`, isError: true };
+          caughtToolErrors.push({ toolName: call.name, message });
+        }
+      }
       io.onToolResult?.(call.name, result);
       const modelOutput = redactSensitiveText(result.output);
       const untrusted = result.untrusted === true;
@@ -59371,7 +59417,7 @@ function isApprovalFor(response, fingerprint) {
   }
   return response.fingerprint === undefined || response.fingerprint === fingerprint;
 }
-async function runConcurrentSpawnBatch(spawnCalls, toolByName, io, deps, hasInvocationCapacity, reserveInvocation, signal) {
+async function runConcurrentSpawnBatch(spawnCalls, toolByName, io, deps, hasInvocationCapacity, reserveInvocation, signal, caughtToolErrors) {
   const maxConcurrency = deps.maxSubagentConcurrency ?? DEFAULT_MAX_SUBAGENT_CONCURRENCY;
   const perTaskRuntimeMs = NOMINAL_CONCURRENT_SPAWN_RUNTIME_MS;
   const tasks = spawnCalls.map((call) => ({
@@ -59398,6 +59444,7 @@ async function runConcurrentSpawnBatch(spawnCalls, toolByName, io, deps, hasInvo
           output: `subagent call ${call.id} failed: sequential fallback error: ${message}`,
           isError: true
         });
+        caughtToolErrors.push({ toolName: call.name, message });
       }
     }
     return results;
@@ -59425,6 +59472,9 @@ async function runConcurrentSpawnBatch(spawnCalls, toolByName, io, deps, hasInvo
     const results = new Map;
     for (const call of spawnCalls) {
       const settled = partialResults?.get(call.id);
+      if (settled === undefined) {
+        caughtToolErrors.push({ toolName: call.name, message });
+      }
       results.set(call.id, settled ?? {
         output: `subagent call ${call.id} failed: concurrent wave error: ${message}`,
         isError: true
@@ -127701,7 +127751,7 @@ async function dispatchLocked(projectRoot, entry, flow, dispatch, deps, runId) {
     });
     const provider = guardUsage(built, () => meter.markUsageMissing());
     try {
-      const tools = buildUnattendedRoster(worktree, unattendedRunner(worktree, sandbox));
+      const tools = [...buildUnattendedRoster(worktree, unattendedRunner(worktree, sandbox)), ...deps.extraTools ?? []];
       const toolNames = tools.map((t) => t.definition.name);
       const shellHooks = buildShellHookRuntime({
         projectRoot: worktree,
@@ -127756,6 +127806,9 @@ async function dispatchLocked(projectRoot, entry, flow, dispatch, deps, runId) {
         signal: controller.signal
       });
       finishReason = result.finishReason;
+      if (result.caughtToolErrors !== undefined && result.caughtToolErrors.length > 0) {
+        crashed = result.caughtToolErrors.map((e) => `${e.toolName}: ${e.message}`).join("; ");
+      }
     } catch (error) {
       crashed = error instanceof Error ? error.message : String(error);
     } finally {
@@ -153368,7 +153421,12 @@ async function runLocked(projectRoot, entry, action, deps, runId, now) {
       const env = deps.grantedEnv ?? process.env;
       const extraSecrets = await ghAuthToken(binaries.realpaths["gh"], grantedCwd, env);
       const granted = action.grants.tools.map((id) => grantedToolSpec(id)).filter((spec) => spec !== undefined).map((spec) => grantedTool(spec, action, grantedCwd, env, grantedCalls, binaries.realpaths, extraSecrets, binaries.stats));
-      const tools = [...builtinReadOnlyTools(projectRoot), shellExecTool(workdir, sandboxedRunner(workdir, sandbox)), ...granted];
+      const tools = [
+        ...builtinReadOnlyTools(projectRoot),
+        shellExecTool(workdir, sandboxedRunner(workdir, sandbox)),
+        ...granted,
+        ...deps.extraTools ?? []
+      ];
       const offending = tools.filter((t) => UNATTENDED_EXCLUDED_TOOLS.includes(t.definition.name));
       if (offending.length > 0)
         throw new Error(`agent-task roster must not offer: ${offending.map((t) => t.definition.name).join(", ")}`);
@@ -153418,6 +153476,9 @@ async function runLocked(projectRoot, entry, action, deps, runId, now) {
       };
       const result = await runAgentTurn(io, agentDeps, history, buildAgentTaskPrompt(entry, action), { signal: controller.signal });
       finishReason = result.finishReason;
+      if (result.caughtToolErrors !== undefined && result.caughtToolErrors.length > 0) {
+        crashed = result.caughtToolErrors.map((e) => `${e.toolName}: ${e.message}`).join("; ");
+      }
     } catch (error) {
       crashed = error instanceof Error ? error.message : String(error);
     } finally {
@@ -155507,14 +155568,19 @@ function attachUsageIo(io, chrome) {
     }
   });
 }
-function estimateTaskCostUsd(profile, inputTokens, outputTokens) {
+var OPENAI_CACHED_INPUT_DISCOUNT = 0.5;
+function estimateTaskCostUsd(profile, providerId, inputTokens, outputTokens, cacheReadTokens) {
   if (profile === undefined)
     return;
   const priceIn = profile.priceInputPerMillion.value;
   const priceOut = profile.priceOutputPerMillion.value;
   if (priceIn === "unknown" || priceOut === "unknown")
     return;
-  return inputTokens / 1e6 * priceIn + outputTokens / 1e6 * priceOut;
+  const cachedInputDiscount = providerId === "openai" || providerId === "openai-codex" ? OPENAI_CACHED_INPUT_DISCOUNT : 1;
+  const cacheRead = cacheReadTokens !== undefined && cacheReadTokens > 0 ? Math.min(cacheReadTokens, inputTokens) : 0;
+  const fullPriceInputTokens = inputTokens - cacheRead;
+  const inputCost = fullPriceInputTokens / 1e6 * priceIn + cacheRead / 1e6 * priceIn * cachedInputDiscount;
+  return inputCost + outputTokens / 1e6 * priceOut;
 }
 async function recordTurnTaskCostBestEffort(input) {
   if (input.inputTokens === 0 && input.outputTokens === 0)
@@ -155522,7 +155588,7 @@ async function recordTurnTaskCostBestEffort(input) {
   try {
     const profiles = loadModelProfiles(input.userConfigDir);
     const profile = profiles[profileKey(input.providerId, input.modelId)];
-    const costUsd = estimateTaskCostUsd(profile, input.inputTokens, input.outputTokens);
+    const costUsd = estimateTaskCostUsd(profile, input.providerId, input.inputTokens, input.outputTokens, input.cacheReadTokens);
     await appendTaskCostRecord({
       providerId: input.providerId,
       modelId: input.modelId,
@@ -160296,10 +160362,12 @@ Usage: /reasoning <${REASONING_EFFORT_LEVELS.join("|")}>
       };
       let turnInputTokens = 0;
       let turnOutputTokens = 0;
+      let turnCacheReadTokens = 0;
       const prevOnUsageForCost = io.onUsage;
       io.onUsage = (usage) => {
         turnInputTokens += usage.inputTokens ?? 0;
         turnOutputTokens += usage.outputTokens ?? 0;
+        turnCacheReadTokens += usage.cacheReadTokens ?? 0;
         prevOnUsageForCost?.(usage);
       };
       const suggestNextStep = async () => {
@@ -160399,6 +160467,7 @@ Usage: /reasoning <${REASONING_EFFORT_LEVELS.join("|")}>
             modelId: deps.modelId,
             inputTokens: turnInputTokens,
             outputTokens: turnOutputTokens,
+            ...turnCacheReadTokens > 0 ? { cacheReadTokens: turnCacheReadTokens } : {},
             success: !turnFailed,
             ...turnCategory !== undefined ? { category: turnCategory } : {}
           });
@@ -162012,7 +162081,9 @@ async function runAgentRepl(lines, rich, deps, metaprojectPort, sessionOpts, ini
       return settled;
     }
     const completion = new Promise((resolve) => {
+      completionWaiters = [];
       completionWaiters.push(() => resolve({ kind: "completion" }));
+      rich.onCompletionWaitersSize?.(completionWaiters.length);
     });
     const winner = await Promise.race([pendingLine, completion]);
     if (winner.kind !== "completion") {
@@ -162677,6 +162748,8 @@ ${GUTTER}${turnSeparator()}
             continue;
           }
           await closeSlateSession(slateSession, mintTimestampAttemptId);
+          lastToolOutput = undefined;
+          lastToolName = undefined;
           if (nextLive !== undefined) {
             live = nextLive;
             history = [];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mrciphersmith/keryx",
-  "version": "0.3.17",
+  "version": "0.3.18",
   "description": "Version-controlled project context for AI coding agents: code graph, architecture wiki, project memory, relevant tests, quality signals, and task flows.",
   "private": false,
   "publishConfig": {