@martinloop/mcp 0.8.3 → 0.8.4

package/dist/package-version.d.ts

@@ -1 +1 @@
-export declare const MARTIN_MCP_PACKAGE_VERSION = "0.8.3";
+export declare const MARTIN_MCP_PACKAGE_VERSION = "0.8.4";

package/dist/package-version.js

@@ -1,3 +1,3 @@
 // Auto-generated by build — do not edit by hand.
 // Source of truth: packages/mcp/package.json#version
-export const MARTIN_MCP_PACKAGE_VERSION = "0.8.3";
+export const MARTIN_MCP_PACKAGE_VERSION = "0.8.4";

package/dist/vendor/adapters/claude-cli.d.ts

@@ -63,6 +63,8 @@ export interface AgentCliAdapterOptions {
     streamingTokenCap?: boolean;
     /** Provider detail-token fields are already included in input/output token totals. */
     streamingUsageDetailsIncludedInTotals?: boolean;
+    /** Whether provider usage arrives incrementally or only on a terminal event. */
+    streamingUsageReportingMode?: "incremental" | "terminal";
     /** Estimated floor used by the host to reject impossible token caps before spawning. */
     budgetPreflight?: {
         minimumViableTokens: number;

package/dist/vendor/adapters/claude-cli.js

@@ -383,7 +383,7 @@ function extractGeminiJsonResult(stdout, modelLabel) {
         })
     };
 }
-function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate, capTokens, detailTokensIncludedInTotals = false, observedUsageGovernor) {
+function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate, capTokens, detailTokensIncludedInTotals = false, observedUsageGovernor, usageReportingMode = "incremental") {
     let pricingResolution = resolveModelPricing(modelLabel);
     // Safety margin: terminate at 80% of cap to bound one-turn overshoot.
     // Without this, a single expensive turn can blow past the cap before the
@@ -424,15 +424,22 @@ function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate,
                 finalObservationPublished = true;
             try {
                 const decision = observedUsageGovernor({ cumulativeUsd, cumulativeTokens, turns, final });
-                if (decision.action === "terminate") {
+                if (decision.action === "terminate" && !(usageReportingMode === "terminal" && final)) {
                     terminate(`Parent observed-usage governor terminated the subprocess: ${decision.reason}.`);
                 }
             }
             catch (error) {
-                terminate(`Parent observed-usage governor failed closed: ${error instanceof Error ? error.message : String(error)}.`);
+                if (!(usageReportingMode === "terminal" && final)) {
+                    terminate(`Parent observed-usage governor failed closed: ${error instanceof Error ? error.message : String(error)}.`);
+                }
             }
             return;
         }
+        // Codex reports authoritative usage on turn.completed, after the provider
+        // turn and its workspace edits have already finished. Record that terminal
+        // settlement, but do not kill and discard completed work at this point.
+        if (usageReportingMode === "terminal" && final)
+            return;
         if (capTokens !== undefined && capTokens > 0 && cumulativeTokens > capTokens) {
             terminate(`Streaming token lease exceeded after ${String(turns)} turn(s): observed ${String(cumulativeTokens)} tokens ` +
                 `surpassed the per-attempt token cap ${String(capTokens)}. Subprocess terminated to prevent additional token usage.`);
@@ -557,7 +564,7 @@ function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate,
             }
             return;
         }
-        extractUsageFromEvent(event, terminate);
+        extractUsageFromEvent(event, terminate, usageReportingMode === "terminal" && event.type === "turn.completed");
     };
     return {
         onChunk: (chunk, terminate) => {
@@ -572,7 +579,8 @@ function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate,
             const bytesCeiling = blendedCostPerChar && blendedCostPerChar > 0
                 ? Math.ceil((capUsd / blendedCostPerChar) * 2)
                 : undefined;
-            if (!usageEventSeen && capUsd > 0 && bytesCeiling !== undefined && totalBytes > bytesCeiling) {
+            if (usageReportingMode === "incremental" &&
+                !usageEventSeen && capUsd > 0 && bytesCeiling !== undefined && totalBytes > bytesCeiling) {
                 terminate(`Streaming byte ceiling exceeded (${String(totalBytes)} bytes > ${String(bytesCeiling)} ceiling) ` +
                     `without any usage events parsed. The Claude stream-json event format may have changed. ` +
                     `Subprocess terminated as a fallback budget guard for cap $${capUsd.toFixed(4)}.`);
@@ -582,7 +590,8 @@ function createStreamingUsageInspector(capUsd, modelLabel, promptTokenEstimate,
             // without a single usage event, estimate spend from byte volume and
             // enforce the cap. This catches cases where Claude's event format
             // changed but bytes are still flowing.
-            if (!usageEventSeen &&
+            if (usageReportingMode === "incremental" &&
+                !usageEventSeen &&
                 capUsd > 0 &&
                 firstChunkAt !== undefined &&
                 Date.now() - firstChunkAt > USAGE_BLIND_TIMEOUT_MS &&
@@ -726,7 +735,7 @@ export function createAgentCliAdapter(options) {
             const streamingUsage = options.streamingUsageCap &&
                 (request.context.remainingBudgetUsd > 0 ||
                     (options.streamingTokenCap && (request.context.remainingTokens ?? 0) > 0))
-                ? createStreamingUsageInspector(request.context.remainingBudgetUsd, options.model ?? options.command, estimatedUsage.tokensIn, options.streamingTokenCap ? request.context.remainingTokens : undefined, options.streamingUsageDetailsIncludedInTotals, request.observedUsageGovernor)
+                ? createStreamingUsageInspector(request.context.remainingBudgetUsd, options.model ?? options.command, estimatedUsage.tokensIn, options.streamingTokenCap ? request.context.remainingTokens : undefined, options.streamingUsageDetailsIncludedInTotals, request.observedUsageGovernor, options.streamingUsageReportingMode)
                 : undefined;
             const agentResult = await runSubprocess(options.executionCommand ?? options.command, args, {
                 cwd: workingDirectory,

package/dist/vendor/adapters/codex-cli.js

@@ -51,6 +51,7 @@ export function createCodexCliAdapter(options = {}) {
         streamingUsageCap: true,
         streamingTokenCap: true,
         streamingUsageDetailsIncludedInTotals: true,
+        streamingUsageReportingMode: "terminal",
         budgetPreflight: {
             minimumViableTokens: CODEX_MINIMUM_VIABLE_TOKEN_BUDGET,
             basis: CODEX_TOKEN_BUDGET_PREFLIGHT_BASIS

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@martinloop/mcp",
-  "version": "0.8.3",
+  "version": "0.8.4",
   "mcpName": "io.github.Keesan12/martin-loop",
   "private": false,
   "type": "module",

package/server.json

@@ -7,12 +7,12 @@
     "url": "https://github.com/Keesan12/martin-loop",
     "source": "github"
   },
-  "version": "0.8.3",
+  "version": "0.8.4",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@martinloop/mcp",
-      "version": "0.8.3",
+      "version": "0.8.4",
       "transport": {
         "type": "stdio"
       }