opencode-anthropic-fix 0.0.38 → 0.0.41

package/lib/backoff.mjs

@@ -7,6 +7,12 @@ const AUTH_FAILED_BACKOFF = 5_000;
 const RATE_LIMIT_EXCEEDED_BACKOFF = 30_000;
 const MIN_BACKOFF_MS = 2_000;
 
+/** Cap on cooldowns derived from reset headers. Prevents lock-out from buggy timestamps. */
+const MAX_COOLDOWN_FROM_RESET_MS = 300_000; // 5 minutes
+
+/** Threshold below which a 429 retry-after is considered transient (burst throttle). */
+export const TRANSIENT_RETRY_THRESHOLD_MS = 10_000; // 10 seconds
+
 /**
  * Parse the retry-after-ms header from a response.
  * Returns milliseconds directly (no conversion needed).
@@ -223,9 +229,9 @@ export function parseRateLimitReason(status, body) {
  * @returns {number}
  */
 export function calculateBackoffMs(reason, consecutiveFailures, retryAfterMs) {
-  // Retry-After header takes precedence
+  // Retry-After header takes precedence, but cap at MAX_COOLDOWN to prevent lock-out
   if (retryAfterMs && retryAfterMs > 0) {
-    return Math.max(retryAfterMs, MIN_BACKOFF_MS);
+    return Math.min(Math.max(retryAfterMs, MIN_BACKOFF_MS), MAX_COOLDOWN_FROM_RESET_MS);
   }
 
   switch (reason) {

package/lib/config.mjs

@@ -86,6 +86,14 @@ import { randomBytes } from "node:crypto";
  * @property {boolean} usage_toast
  * @property {CCCredentialReuseConfig} cc_credential_reuse
  * @property {AdaptiveContextConfig} adaptive_context
+ * @property {{ enabled: boolean, default_max_tokens: number, escalated_max_tokens: number }} output_cap
+ * @property {{ enabled: boolean, timeout_ms: number }} preconnect
+ * @property {{ enabled: boolean, safety_margin: number }} overflow_recovery
+ * @property {{ enabled: boolean, alert_threshold: number }} cache_break_detection
+ * @property {{ enabled: boolean, background_max_service_retries: number, background_max_should_retries: number }} request_classification
+ * @property {{ enabled: boolean, default: number, completion_threshold: number }} token_budget
+ * @property {{ enabled: boolean, threshold_percent: number }} microcompact
+ * @property {{ enabled: boolean, default_cooldown_ms: number, poll_quota_on_overload: boolean }} overload_recovery
  */
 
 /** @type {AnthropicAuthConfig} */
@@ -168,16 +176,62 @@ export const DEFAULT_CONFIG = {
   },
   /** Token economy optimizations. */
   token_economy: {
-    /** Enable token-efficient-tools-2026-03-28 beta (~4.5% output token reduction).
-     *  Mutually exclusive with structured-outputs — this flag takes precedence. */
-    token_efficient_tools: true,
+    /** [DEPRECATED] token-efficient-tools-2026-03-28 was removed in CC v2.1.90.
+     *  This flag is retained for config backward compatibility but has no effect. */
+    token_efficient_tools: false,
     /** Enable redact-thinking-2026-02-12 beta (suppresses thinking summaries
-     *  server-side for lower bandwidth). Off by default — set to true if you
+     *  server-side for lower bandwidth). Off by default u2014 set to true if you
      *  don't need to inspect thinking blocks. */
     redact_thinking: false,
-    /** Enable summarize-connector-text-2026-03-13 beta (API summarizes assistant
-     *  text between tool calls — anti-distillation measure). */
-    connector_text_summarization: true,
+    /** [DEPRECATED] summarize-connector-text-2026-03-13 was removed in CC v2.1.90.
+     *  This flag is retained for config backward compatibility but has no effect. */
+    connector_text_summarization: false,
+  },
+  /** Output cap: default max_tokens to save context window. */
+  output_cap: {
+    enabled: true,
+    default_max_tokens: 8_000,
+    escalated_max_tokens: 64_000,
+  },
+  /** API preconnect: fire-and-forget HEAD to pre-warm TCP+TLS. */
+  preconnect: {
+    enabled: true,
+    timeout_ms: 10_000,
+  },
+  /** Context overflow auto-recovery: parse error, reduce max_tokens, retry once. */
+  overflow_recovery: {
+    enabled: true,
+    safety_margin: 1_000,
+  },
+  /** Cache break detection: alert when cache_read_input_tokens drops significantly. */
+  cache_break_detection: {
+    enabled: true,
+    alert_threshold: 2_000,
+  },
+  /** Request classification: reduced retry budget for background requests. */
+  request_classification: {
+    enabled: true,
+    background_max_service_retries: 0,
+    background_max_should_retries: 1,
+  },
+  /** Token budget: parse NL budget expressions, track output, enforce limits. */
+  token_budget: {
+    enabled: false,
+    default: 0,
+    completion_threshold: 0.9,
+  },
+  /** Microcompact: inject clear_tool_uses/clear_thinking betas at high context utilization. */
+  microcompact: {
+    enabled: true,
+    threshold_percent: 80,
+  },
+  /** Overload recovery: quota-aware account switching on 529 exhaustion. */
+  overload_recovery: {
+    enabled: true,
+    /** Cooldown (ms) applied to overloaded accounts when no quota data available. */
+    default_cooldown_ms: 60_000,
+    /** Whether to poll /api/oauth/usage on 529 exhaustion for smarter cooldowns. */
+    poll_quota_on_overload: true,
   },
 };
 
@@ -208,6 +262,14 @@ function createDefaultConfig() {
     adaptive_context: { ...DEFAULT_CONFIG.adaptive_context },
     willow_mode: { ...DEFAULT_CONFIG.willow_mode },
     token_economy: { ...DEFAULT_CONFIG.token_economy },
+    output_cap: { ...DEFAULT_CONFIG.output_cap },
+    preconnect: { ...DEFAULT_CONFIG.preconnect },
+    overflow_recovery: { ...DEFAULT_CONFIG.overflow_recovery },
+    cache_break_detection: { ...DEFAULT_CONFIG.cache_break_detection },
+    request_classification: { ...DEFAULT_CONFIG.request_classification },
+    token_budget: { ...DEFAULT_CONFIG.token_budget },
+    microcompact: { ...DEFAULT_CONFIG.microcompact },
+    overload_recovery: { ...DEFAULT_CONFIG.overload_recovery },
   };
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-anthropic-fix",
-  "version": "0.0.38",
+  "version": "0.0.41",
   "license": "GPL-3.0-or-later",
   "main": "./index.mjs",
   "files": [