fix(proxy): downstream abort propagation, Anthropic stream completion, history filtering

- Cancel upstream requests when client disconnects via abort signal
- Stop reading Anthropic SSE after message_stop event
- Skip inference history for non-LLM requests and empty records
- Use Intl.NumberFormat for full USDC precision (up to 12 decimals)
- Guard stream res.end() against double-close
- Default claude model changed to stepfun/step-3.5-flash

Author: tenequm

packages/x402-proxy/CHANGELOG.md

@@ -7,6 +7,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.10.6] - 2026-04-01
+
+### Fixed
+- Client disconnect now cancels upstream requests via abort signal propagation (prevents resource leaks on dropped connections)
+- Anthropic SSE streaming stops reading after `message_stop` event instead of waiting for connection close
+- Non-LLM endpoint requests (tool discovery, resource listing) no longer write empty records to inference history
+- Empty inference history records (no amount, model, tokens, or tx) filtered out when reading history
+- `formatUsdcValue()` uses `Intl.NumberFormat` for full precision up to 12 decimals instead of truncating to fixed tiers
+- Stream responses guarded against double `res.end()` calls
+
+### Changed
+- Default model for `claude` command changed from `minimax/minimax-m2.7` to `stepfun/step-3.5-flash`
+
 ## [0.10.5] - 2026-04-01
 
 ### Fixed
@@ -374,7 +387,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `appendHistory` / `readHistory` / `calcSpend` - JSONL transaction history
 - Re-exports from `@x402/fetch`, `@x402/svm`, `@x402/evm`
 
-[Unreleased]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.5...HEAD
+[Unreleased]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.6...HEAD
+[0.10.6]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.5...v0.10.6
 [0.10.5]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.4...v0.10.5
 [0.10.4]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.3...v0.10.4
 [0.10.3]: https://github.com/cascade-protocol/x402-proxy/compare/v0.10.2...v0.10.3

packages/x402-proxy/package.json

@@ -1,6 +1,6 @@
 {
   "name": "x402-proxy",
-  "version": "0.10.5",
+  "version": "0.10.6",
   "description": "curl for x402 paid APIs. Auto-pays any endpoint on Base, Solana, and Tempo. Also works as an OpenClaw plugin.",
   "type": "module",
   "sideEffects": false,

packages/x402-proxy/src/commands/claude.ts

@@ -4,7 +4,7 @@ import { dim, error } from "../lib/output.js";
 import { DEFAULT_SURF_UPSTREAM_URL } from "../openclaw/defaults.js";
 import { startServeServer } from "./serve.js";
 
-const DEFAULT_MODEL = "minimax/minimax-m2.7";
+const DEFAULT_MODEL = "stepfun/step-3.5-flash";
 
 type ClaudeFlags = {
   model: string;
@@ -34,7 +34,7 @@ Examples:
     flags: {
       model: {
         kind: "parsed",
-        brief: "Model to use (default: minimax/minimax-m2.7)",
+        brief: "Model to use (default: stepfun/step-3.5-flash)",
         parse: String,
         default: DEFAULT_MODEL,
       },

packages/x402-proxy/src/history.ts

@@ -45,6 +45,18 @@ export type TxRecord = {
   meta?: Record<string, string | number>;
 };
 
+function isMeaningfulInferenceRecord(record: TxRecord): boolean {
+  if (record.kind !== "x402_inference") return true;
+  if (!record.ok) return true;
+  return (
+    record.amount != null ||
+    record.model != null ||
+    record.inputTokens != null ||
+    record.outputTokens != null ||
+    record.tx != null
+  );
+}
+
 // --- File operations ---
 
 export function appendHistory(historyPath: string, record: TxRecord): void {
@@ -74,7 +86,8 @@ export function readHistory(historyPath: string): TxRecord[] {
       try {
         const parsed = JSON.parse(line);
         if (typeof parsed.t !== "number" || typeof parsed.kind !== "string") return [];
-        return [parsed as TxRecord];
+        const record = parsed as TxRecord;
+        return isMeaningfulInferenceRecord(record) ? [record] : [];
       } catch {
         return [];
       }
@@ -109,12 +122,12 @@ export function calcSpend(records: TxRecord[]): {
 
 // --- Formatting ---
 
-/** Format a USDC value with adaptive precision (no token suffix). */
 export function formatUsdcValue(amount: number): string {
-  if (amount >= 0.01) return amount.toFixed(2);
-  if (amount >= 0.001) return amount.toFixed(3);
-  if (amount >= 0.0001) return amount.toFixed(4);
-  return amount.toFixed(6);
+  return new Intl.NumberFormat("en-US", {
+    useGrouping: false,
+    minimumFractionDigits: 0,
+    maximumFractionDigits: 12,
+  }).format(amount);
 }
 
 export function formatAmount(amount: number, token: string): string {

packages/x402-proxy/src/openclaw/route.test.ts

@@ -1,6 +1,6 @@
 import type { ServerResponse } from "node:http";
 import { describe, expect, it } from "vitest";
-import { createSseTracker, writeErrorResponse } from "./route.js";
+import { createSseTracker, shouldAppendInferenceHistory, writeErrorResponse } from "./route.js";
 
 // ── createSseTracker ────────────────────────────────────────────────
 
@@ -125,6 +125,7 @@ describe("createSseTracker", () => {
         cacheRead: undefined,
         cacheWrite: undefined,
       });
+      expect(tracker.sawAnthropicMessageStop).toBe(true);
     });
 
     it("extracts cache fields from message_start", () => {
@@ -189,6 +190,7 @@ describe("createSseTracker", () => {
         cacheRead: undefined,
         cacheWrite: undefined,
       });
+      expect(tracker.sawAnthropicMessageStop).toBe(true);
     });
   });
 
@@ -293,6 +295,23 @@ describe("createSseTracker", () => {
   });
 });
 
+describe("shouldAppendInferenceHistory", () => {
+  it("skips non-LLM traffic and empty successful placeholders", () => {
+    expect(shouldAppendInferenceHistory({ isLlmEndpoint: false })).toBe(false);
+    expect(shouldAppendInferenceHistory({ isLlmEndpoint: true })).toBe(false);
+  });
+
+  it("keeps usage-bearing and priced LLM requests", () => {
+    expect(shouldAppendInferenceHistory({ isLlmEndpoint: true, amount: 0.0133 })).toBe(true);
+    expect(
+      shouldAppendInferenceHistory({
+        isLlmEndpoint: true,
+        usage: { model: "stepfun/step-3.5-flash", inputTokens: 10, outputTokens: 2 },
+      }),
+    ).toBe(true);
+  });
+});
+
 // ── writeErrorResponse ──────────────────────────────────────────────
 
 describe("writeErrorResponse", () => {