codex-multi-auth 2.9.2 → 2.11.0

package/dist/lib/request/error-classification.js

@@ -11,6 +11,38 @@ const CHATGPT_CODEX_UNSUPPORTED_MODEL_PATTERN = /model is not supported when usi
 const NORMALIZED_UNSUPPORTED_MODEL_PATTERN = /the model ['"]([^'"]+)['"] is not currently available for this chatgpt account/i;
 const MODEL_ACCESS_DENIED_PATTERN = /the model [`'"]([^`'"]+)[`'"] does not exist or you do not have access to it/i;
 export const DEFAULT_UNSUPPORTED_CODEX_FALLBACK_CHAIN = {
+    // GPT-6 Astra rolls out org by org, so an account that is not entitled yet
+    // gets a real unsupported-model response for it. This chain only fires when
+    // the user has opted into `fallbackOnUnsupportedCodexModel` (default
+    // `false`), and only on that response, so it never silently swaps the model
+    // out from under a request the account could have served.
+    //
+    // This table is walked ONE HOP AT A TIME, not as a per-request candidate
+    // list. index.ts reassigns `model` to whatever came back and passes that as
+    // `requestedModel` on the next unsupported response, so reaching `gpt-5.5`
+    // from Astra requires `gpt-5.6-sol` to carry its own entry below. A second
+    // element here is only ever consulted when the first is already in
+    // `attemptedModels` for that same call.
+    //
+    // Depth is not free: every hop spends one of the shared per-request outbound
+    // attempts (`tryConsumeOutboundRequestAttempt`). A single-account balanced
+    // session gets a budget of 5, and the aeon walk needs exactly 5, so it fits
+    // with nothing spare. Spend an attempt on a retry or a stream failover and
+    // the tail hops become unreachable, ending as an attempt-budget-exhausted
+    // 503 rather than `gpt-5.4`. Hops are ordered most-valuable-first so what is
+    // lost first matters least. Do not add a hop to a GPT-6 row without
+    // re-checking that budget; `test/gpt6-astra-models.test.ts` asserts it.
+    "gpt-6-astra": ["gpt-5.6-sol", "gpt-5.5"],
+    // aeon steps to the flagship first: still GPT-6, still Astra, just without
+    // the long-horizon behaviour.
+    "gpt-6-astra-aeon": ["gpt-6-astra", "gpt-5.6-sol"],
+    // The hop that makes the Astra path reach a model every account has. GPT-5.6
+    // shipped without a chain entry, so an unsupported 5.6 response ended the
+    // walk; with Astra above it that would have stranded the fallback one rung
+    // short of the floor it documents. Terra and Luna are deliberately still
+    // absent: nothing steps into them, so giving them a hop would change 5.6
+    // behaviour beyond completing this path.
+    "gpt-5.6-sol": ["gpt-5.5"],
     "gpt-5": ["gpt-5.5"],
     "gpt-5-pro": ["gpt-5.5-pro"],
     "gpt-5-chat-latest": ["gpt-5.5"],
@@ -43,7 +75,14 @@ function canonicalizeModelName(model) {
     return stripped.replace(/-(none|minimal|low|medium|high|xhigh)$/i, "");
 }
 function normalizeFallbackChain(customChain) {
-    const normalized = {};
+    // Null-prototype, because both the keys written here and the keys read out of
+    // it are caller-controlled. On a plain object, a `customChain` entry named
+    // `__proto__` reassigns this object's prototype instead of adding a row, and
+    // a lookup for `constructor` returns `Object.prototype.constructor`, a
+    // truthy non-array whose `.length` is 1, so the emptiness check passes and
+    // the `for...of` below throws `targets is not iterable` inside the request
+    // path.
+    const normalized = Object.create(null);
     for (const [key, values] of Object.entries(DEFAULT_UNSUPPORTED_CODEX_FALLBACK_CHAIN)) {
         const normalizedKey = canonicalizeModelName(key);
         if (!normalizedKey)
@@ -158,7 +197,9 @@ export function resolveUnsupportedCodexFallbackModel(options) {
     }
     const chain = normalizeFallbackChain(options.customChain);
     const targets = chain[currentModel] ?? [];
-    if (targets.length === 0)
+    // Belt and braces with the null-prototype chain above: a `customChain` the
+    // caller passed straight through could still carry a non-array value.
+    if (!Array.isArray(targets) || targets.length === 0)
         return undefined;
     for (const target of targets) {
         if (!options.fallbackToGpt52OnUnsupportedGpt53 &&

package/dist/lib/request/helpers/model-map.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"model-map.d.ts","sourceRoot":"","sources":["../../../../lib/request/helpers/model-map.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAKH,OAAO,KAAK,EACX,oBAAoB,EACpB,mBAAmB,EACnB,MAAM,oBAAoB,CAAC;AAE5B,YAAY,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,CAAC;AAE1D,MAAM,MAAM,iBAAiB,GAC1B,aAAa,GACb,WAAW,GACX,OAAO,GACP,SAAS,GACT,SAAS,CAAC;AAEb;;;GAGG;AACH,MAAM,MAAM,WAAW,GAAG,iBAAiB,CAAC;AAE5C;;;GAGG;AACH,eAAO,MAAM,cAAc,EAAE,SAAS,WAAW,EAMvC,CAAC;AAEX,MAAM,WAAW,iBAAiB;IACjC,UAAU,EAAE,OAAO,CAAC;IACpB,WAAW,EAAE,OAAO,CAAC;IACrB,UAAU,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,YAAY;IAC5B,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,iBAAiB,CAAC;IAChC,sBAAsB,EAAE,oBAAoB,CAAC;IAC7C,yBAAyB,EAAE,SAAS,oBAAoB,EAAE,CAAC;IAC3D,YAAY,EAAE,iBAAiB,CAAC;CAChC;AA6CD,eAAO,MAAM,mBAAmB,kBAAkB,CAAC;AACnD,eAAO,MAAM,aAAa,YAAY,CAAC;AAQvC,eAAO,MAAM,mBAAmB,gBAAgB,CAAC;AAOjD,eAAO,MAAM,uBAAuB,iGAO1B,CAAC;AA4EX;;;;;;;GAOG;AACH,eAAO,MAAM,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CA6G9C,CAAC;AAEX,QAAA,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAM,CAAC;AAqG7C,OAAO,EAAE,SAAS,EAAE,CAAC;AAoKrB;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAQtE;AAED;;;;;;GAMG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CA2BxE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,YAAY,CAavE;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,iBAAiB,CAEjF;AAeD;;;;;;;;;;;GAWG;AACH,wBAAgB,2BAA2B,CAC1C,KAAK,EAAE,MAAM,GAAG,SAAS,GACvB,mBAAmB,CASrB;AAED;;;;;GAKG;AACH,wBAAgB,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAErD"}
+{"version":3,"file":"model-map.d.ts","sourceRoot":"","sources":["../../../../lib/request/helpers/model-map.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAKH,OAAO,KAAK,EACX,oBAAoB,EACpB,mBAAmB,EACnB,MAAM,oBAAoB,CAAC;AAE5B,YAAY,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,CAAC;AAE1D,MAAM,MAAM,iBAAiB,GAC1B,aAAa,GACb,WAAW,GACX,OAAO,GACP,SAAS,GACT,SAAS,CAAC;AAEb;;;GAGG;AACH,MAAM,MAAM,WAAW,GAAG,iBAAiB,CAAC;AAE5C;;;GAGG;AACH,eAAO,MAAM,cAAc,EAAE,SAAS,WAAW,EAMvC,CAAC;AAEX,MAAM,WAAW,iBAAiB;IACjC,UAAU,EAAE,OAAO,CAAC;IACpB,WAAW,EAAE,OAAO,CAAC;IACrB,UAAU,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,YAAY;IAC5B,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,iBAAiB,CAAC;IAChC,sBAAsB,EAAE,oBAAoB,CAAC;IAC7C,yBAAyB,EAAE,SAAS,oBAAoB,EAAE,CAAC;IAC3D,YAAY,EAAE,iBAAiB,CAAC;CAChC;AA6CD,eAAO,MAAM,mBAAmB,kBAAkB,CAAC;AACnD,eAAO,MAAM,aAAa,YAAY,CAAC;AAevC,eAAO,MAAM,mBAAmB,gBAAgB,CAAC;AAOjD,eAAO,MAAM,uBAAuB,iGAO1B,CAAC;AAoIX;;;;;;;GAOG;AACH,eAAO,MAAM,cAAc,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CA+I9C,CAAC;AAEX,QAAA,MAAM,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAM,CAAC;AAgI7C,OAAO,EAAE,SAAS,EAAE,CAAC;AAqOrB;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAQtE;AAED;;;;;;GAMG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,CAuCxE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,YAAY,CAavE;AAED;;GAEG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,iBAAiB,CAEjF;AAeD;;;;;;;;;;;GAWG;AACH,wBAAgB,2BAA2B,CAC1C,KAAK,EAAE,MAAM,GAAG,SAAS,GACvB,mBAAmB,CASrB;AAED;;;;;GAKG;AACH,wBAAgB,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAErD"}

package/dist/lib/request/helpers/model-map.js

@@ -60,6 +60,13 @@ export const DEFAULT_MODEL = "gpt-5.5";
 // actual request routing and the legacy `gpt-5` alias remain opt-in per 2.5.0.
 // Bare `gpt-5.6` aliases to Sol; we pin the canonical id so the probe display
 // and report `modelSelection` read `gpt-5.6-sol` without a remap arrow.
+//
+// GPT-6 Astra deliberately does NOT lead the probe as of its 2026-09-03 launch.
+// A probe only needs a response's quota headers, and Astra is still rolling out
+// org by org, so leading with it would spend one failed request per probe for
+// every account without entitlement yet and buy nothing. Move DEFAULT_PROBE_MODEL
+// to `gpt-6-astra` once it is broadly available, and put `gpt-5.6-sol` directly
+// behind it in QUOTA_PROBE_MODEL_CHAIN.
 export const DEFAULT_PROBE_MODEL = "gpt-5.6-sol";
 // Single source of truth for the live/quota probe fallback chain. Both the
 // manager probe (lib/quota-probe.ts) and the runtime probe (lib/runtime/quota-probe.ts)
@@ -102,6 +109,57 @@ const GPT_5_6_LUNA_EFFORTS = [
     "xhigh",
     "max",
 ];
+/**
+ * GPT-6 Astra, OpenAI's 2026-09-03 frontier release.
+ *
+ * There is no Sol/Terra/Luna split this generation: the announced lineup is the
+ * flagship plus `aeon`, a long-horizon variant built for runs measured in days.
+ * `gpt-6-astra` is the API model name OpenAI published at launch;
+ * `gpt-6-astra-aeon` is the second slug that shipped beside it in the Codex
+ * model list. "Astra Pro" is a plan tier, not a separate slug we have seen, so
+ * it is deliberately not registered as its own canonical model — the GPT-6
+ * resolver below claims `gpt-6-astra-pro` and every other unrecognised GPT-6 id
+ * for the flagship rather than letting it fall through to GPT-5.5.
+ */
+const GPT_6_ASTRA_MODEL = "gpt-6-astra";
+const GPT_6_ASTRA_AEON_MODEL = "gpt-6-astra-aeon";
+/** Bare `gpt-6` resolves to the flagship, mirroring bare `gpt-5.6` -> Sol. */
+const GPT_6_FLAGSHIP_ALIAS = "gpt-6";
+/**
+ * Astra inherits the GPT-5.6 frontier effort ladder: no `none`/`minimal`, and
+ * `ultra` at the top. OpenAI has published no default effort for Astra, so the
+ * flagship follows the tier it succeeds (Sol, `low`) and the long-horizon
+ * variant follows the catalog's other long-running models (`medium`).
+ */
+const GPT_6_ASTRA_EFFORTS = [
+    "low",
+    "medium",
+    "high",
+    "xhigh",
+    "max",
+    "ultra",
+];
+/**
+ * Cyber-specialty models from the upstream Codex catalog
+ * (openai/codex `codex-rs/models-manager/models.json`).
+ *
+ * They are hidden in the Codex picker and gated behind the Daybreak program,
+ * but the catalog marks them `supported_in_api`, so a client can and does name
+ * them. Until now every `gpt-daybreak-*` id missed the codex resolver (no
+ * `codex` token) and the general GPT-5 resolver (no `gpt 5` tokens) and landed
+ * on `DEFAULT_MODEL` — asking for the cyber-permissive model silently ran
+ * GPT-5.5. Reasoning ladders and defaults below are the catalog's own values.
+ */
+const DAYBREAK_BLUE_MODEL = "gpt-daybreak-blue-latest";
+const DAYBREAK_RED_MODEL = "gpt-daybreak-red-latest";
+const DAYBREAK_EFFORTS = [
+    "low",
+    "medium",
+    "high",
+    "xhigh",
+    "max",
+    "ultra",
+];
 const GPT_5_5_CANONICAL_MODEL = "gpt-5.5";
 const GPT_5_5_PRO_CANONICAL_MODEL = "gpt-5.5-pro";
 const GPT_5_5_RELEASE_MODEL = "gpt-5.5-2026-04-23";
@@ -180,6 +238,40 @@ export const MODEL_PROFILES = {
         supportedReasoningEfforts: ["medium"],
         capabilities: TOOL_CAPABILITIES.compactOnly,
     },
+    // Like GPT-5.6, GPT-6 Astra ships its base instructions inline in the
+    // upstream model catalog rather than as a `gpt_6_prompt.md`, so it stays on
+    // the GPT-5.2 prompt family with every other post-5.2 general model. Adding
+    // a `gpt-6` prompt family here would also widen MODEL_FAMILIES, which is a
+    // persisted key space (`activeIndexByFamily`) and would need a storage
+    // migration to grow.
+    [GPT_6_ASTRA_MODEL]: {
+        normalizedModel: GPT_6_ASTRA_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "low",
+        supportedReasoningEfforts: GPT_6_ASTRA_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
+    [GPT_6_ASTRA_AEON_MODEL]: {
+        normalizedModel: GPT_6_ASTRA_AEON_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "medium",
+        supportedReasoningEfforts: GPT_6_ASTRA_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
+    [DAYBREAK_BLUE_MODEL]: {
+        normalizedModel: DAYBREAK_BLUE_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "low",
+        supportedReasoningEfforts: DAYBREAK_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
+    [DAYBREAK_RED_MODEL]: {
+        normalizedModel: DAYBREAK_RED_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "medium",
+        supportedReasoningEfforts: DAYBREAK_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
     // GPT-5.6 ships its base instructions inline in the upstream model catalog
     // rather than as a `gpt_5_6_prompt.md`, so these stay on the GPT-5.2 prompt
     // family alongside the other post-5.2 general models.
@@ -282,6 +374,21 @@ function addGpt56Aliases() {
     addEffortAliases(GPT_5_6_LUNA_MODEL, GPT_5_6_LUNA_MODEL, GPT_5_6_LUNA_EFFORTS);
     addEffortAliases(GPT_5_6_FLAGSHIP_ALIAS, GPT_5_6_SOL_MODEL, GPT_5_6_SOL_TERRA_EFFORTS);
 }
+function addGpt6Aliases() {
+    addEffortAliases(GPT_6_ASTRA_MODEL, GPT_6_ASTRA_MODEL, GPT_6_ASTRA_EFFORTS);
+    addEffortAliases(GPT_6_ASTRA_AEON_MODEL, GPT_6_ASTRA_AEON_MODEL, GPT_6_ASTRA_EFFORTS);
+    addEffortAliases(GPT_6_FLAGSHIP_ALIAS, GPT_6_ASTRA_MODEL, GPT_6_ASTRA_EFFORTS);
+    // `astra` on its own is how the model is spoken about everywhere; accept it
+    // rather than letting it fall through to GPT-5.5.
+    addEffortAliases("astra", GPT_6_ASTRA_MODEL, GPT_6_ASTRA_EFFORTS);
+    addEffortAliases("astra-aeon", GPT_6_ASTRA_AEON_MODEL, GPT_6_ASTRA_EFFORTS);
+}
+function addDaybreakAliases() {
+    addEffortAliases(DAYBREAK_BLUE_MODEL, DAYBREAK_BLUE_MODEL, DAYBREAK_EFFORTS);
+    addEffortAliases(DAYBREAK_RED_MODEL, DAYBREAK_RED_MODEL, DAYBREAK_EFFORTS);
+    addEffortAliases("daybreak-blue", DAYBREAK_BLUE_MODEL, DAYBREAK_EFFORTS);
+    addEffortAliases("daybreak-red", DAYBREAK_RED_MODEL, DAYBREAK_EFFORTS);
+}
 function addGeneralAliases() {
     addReasoningAliases(GPT_5_5_CANONICAL_MODEL, GPT_5_5_CANONICAL_MODEL);
     addReasoningAliases(GPT_5_5_RELEASE_MODEL, GPT_5_5_CANONICAL_MODEL);
@@ -320,6 +427,8 @@ function addCodexAliases() {
 addCodexAliases();
 addGeneralAliases();
 addGpt56Aliases();
+addGpt6Aliases();
+addDaybreakAliases();
 export { MODEL_MAP };
 function stripProviderPrefix(modelId) {
     return modelId.includes("/") ? (modelId.split("/").pop() ?? modelId) : modelId;
@@ -395,6 +504,68 @@ function resolveCodexCatalogModel(modelId) {
  * model than the caller asked for. Unrecognised tiers resolve to Sol, matching
  * OpenAI's bare `gpt-5.6` alias.
  */
+/**
+ * Resolve GPT-6 identifiers that are not exact aliases — a dated snapshot id
+ * (`gpt-6-astra-2026-09-03`), the "Astra Pro" plan tier (`gpt-6-astra-pro`), or
+ * any tier name OpenAI adds after this file was written.
+ *
+ * This is the same guard the 5.6 resolver exists for, one major version up:
+ * without it, `resolveGeneralGpt5CatalogModel` never matches (it requires a
+ * `gpt 5` token pair) and every unrecognised GPT-6 id lands on `DEFAULT_MODEL`,
+ * running GPT-5.5 for a caller who asked for the frontier model. `aeon` keeps
+ * its own canonical id because it is a behaviourally different model (long
+ * horizon), not a rename of the flagship; everything else resolves to the
+ * flagship, matching OpenAI's bare `gpt-6` alias.
+ *
+ * Ids carrying a `codex` token are left to `resolveCodexCatalogModel`, exactly
+ * as the 5.6 resolver defers them.
+ */
+function resolveGpt6CatalogModel(modelId) {
+    const tokens = tokenizeModelId(modelId);
+    const gptIndex = tokens.indexOf("gpt");
+    const versionToken = gptIndex === -1 ? undefined : tokens[gptIndex + 1];
+    // `gpt6` with no separator tokenizes as ONE token, so the `gpt` + `6` pair
+    // never forms and it used to fall through to GPT-5.5 while still passing
+    // capability-policy's catalog gate. Gate and resolver have to agree on the
+    // same id or the policy store keys state a request never reads.
+    const isGpt6 = versionToken === "6" || tokens.includes("gpt6");
+    // A bare `astra` token counts too. OpenAI's own launch material and every
+    // picker label say "Astra" without the `gpt-6` prefix, so `Astra Pro` and
+    // `astra-fast` reach this resolver with no version tokens at all; without
+    // this clause they miss every branch and land on GPT-5.5.
+    //
+    // It is anchored, not a free-floating substring: an id that names a
+    // DIFFERENT GPT major version does not get claimed for the frontier model
+    // just because `astra` appears in it, so `gpt-4-astra-x` is declined here
+    // rather than silently running GPT-6.
+    const namesOtherGptVersion = versionToken !== undefined &&
+        /^\d+$/.test(versionToken) &&
+        versionToken !== "6";
+    const isAstra = tokens.includes("astra") && !namesOtherGptVersion;
+    if ((!isGpt6 && !isAstra) || tokens.includes("codex")) {
+        return undefined;
+    }
+    if (tokens.includes("aeon"))
+        return GPT_6_ASTRA_AEON_MODEL;
+    return GPT_6_ASTRA_MODEL;
+}
+/**
+ * Resolve the Daybreak cyber models, including ids that are not exact aliases
+ * (a pinned `gpt-daybreak-red-2026-08-14`, say).
+ *
+ * `red` is the cyber-permissive variant and `blue` the defensive one; an
+ * unrecognised Daybreak id resolves to `blue`, the more restricted of the two,
+ * so a typo cannot silently upgrade a caller into the permissive model.
+ */
+function resolveDaybreakCatalogModel(modelId) {
+    const tokens = tokenizeModelId(modelId);
+    if (!tokens.includes("daybreak")) {
+        return undefined;
+    }
+    if (tokens.includes("red"))
+        return DAYBREAK_RED_MODEL;
+    return DAYBREAK_BLUE_MODEL;
+}
 function resolveGpt56CatalogModel(modelId) {
     const tokens = tokenizeModelId(modelId);
     const gptIndex = tokens.indexOf("gpt");
@@ -476,10 +647,20 @@ export function resolveNormalizedModel(model) {
     if (mappedModel) {
         return mappedModel;
     }
+    // Daybreak first: its slugs carry neither a `codex` nor a `gpt 5` token, so
+    // every other resolver declines them and they would reach DEFAULT_MODEL.
+    const daybreakCatalogModel = resolveDaybreakCatalogModel(modelId);
+    if (daybreakCatalogModel) {
+        return daybreakCatalogModel;
+    }
     const codexCatalogModel = resolveCodexCatalogModel(modelId);
     if (codexCatalogModel) {
         return codexCatalogModel;
     }
+    const gpt6CatalogModel = resolveGpt6CatalogModel(modelId);
+    if (gpt6CatalogModel) {
+        return gpt6CatalogModel;
+    }
     const gpt56CatalogModel = resolveGpt56CatalogModel(modelId);
     if (gpt56CatalogModel) {
         return gpt56CatalogModel;

package/dist/lib/request/rate-limit-decision.d.ts

@@ -60,11 +60,10 @@ export interface PinnedUnavailableContext {
     recoveryBound?: "rate-limit" | "other" | "unknown";
     /**
      * The pin's CURRENT runtime skip reason, re-read from account state at the
-     * moment the 503 is built. `accountSkipReasons` holds the SELECTION
-     * verdict, and on the dominant path the retry loop has already overwritten
-     * it with its own bookkeeping token ("already-attempted") — wording the
-     * sentence from that is what issue #675 reported. Omitted when the caller
-     * did not re-read runtime state.
+     * moment the 503 is built. `accountSkipReasons` holds the most recent
+     * selection or attempt verdict, which can be less specific than a cooldown
+     * or breaker created later in the same pass. Omitted when the caller did not
+     * re-read runtime state.
      */
     currentSkipReason?: string | null;
     now?: number;

package/dist/lib/request/rate-limit-decision.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"rate-limit-decision.d.ts","sourceRoot":"","sources":["../../../lib/request/rate-limit-decision.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,qCAAqC,CAAC;AAE5E,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAe/C,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAGlE;AAED,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,OAAO,CAAC,WAAW,EAAE;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,CAAC,GAAG,OAAO,CAWjG;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAapF;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAoBlF;AAYD,wBAAgB,0BAA0B,CAAC,gBAAgB,EAAE,MAAM,GAAG,MAAM,CAuB3E;AAED,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAkBxE;AA0BD,wBAAgB,4BAA4B,CAC3C,OAAO,EAAE,OAAO,EAChB,kBAAkB,EAAE,MAAM,EAC1B,GAAG,EAAE,MAAM,GACT,MAAM,CAcR;AAED,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM,CAE1E;AAED;;;;;;GAMG;AACH,MAAM,WAAW,0BAA0B;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,kCAAkC,CAAC;IACzC,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,mEAAmE;IACnE,UAAU,EAAE,QAAQ,GAAG,QAAQ,GAAG,IAAI,CAAC;IACvC,2EAA2E;IAC3E,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC7C;AAQD,MAAM,WAAW,wBAAwB;IACxC;;;;;;OAMG;IACH,SAAS,CAAC,EAAE,QAAQ,GAAG,QAAQ,GAAG,IAAI,CAAC;IACvC,uEAAuE;IACvE,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B;;;;;;;;;;;;;;;;OAgBG;IACH,aAAa,CAAC,EAAE,YAAY,GAAG,OAAO,GAAG,SAAS,CAAC;IACnD;;;;;;;OAOG;IACH,iBAAiB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,GAAG,CAAC,EAAE,MAAM,CAAC;CACb;AAwID,wBAAgB,+BAA+B,CAC9C,WAAW,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,EACtC,kBAAkB,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,EAC/C,OAAO,CAAC,EAAE,wBAAwB,GAChC,0BAA0B,CA8E5B"}
+{"version":3,"file":"rate-limit-decision.d.ts","sourceRoot":"","sources":["../../../lib/request/rate-limit-decision.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,qCAAqC,CAAC;AAE5E,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAe/C,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAGlE;AAED,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,OAAO,CAAC,WAAW,EAAE;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,CAAC,GAAG,OAAO,CAWjG;AAED,wBAAgB,uBAAuB,CAAC,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAapF;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAoBlF;AAYD,wBAAgB,0BAA0B,CAAC,gBAAgB,EAAE,MAAM,GAAG,MAAM,CAuB3E;AAED,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAkBxE;AA0BD,wBAAgB,4BAA4B,CAC3C,OAAO,EAAE,OAAO,EAChB,kBAAkB,EAAE,MAAM,EAC1B,GAAG,EAAE,MAAM,GACT,MAAM,CAcR;AAED,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM,CAE1E;AAED;;;;;;GAMG;AACH,MAAM,WAAW,0BAA0B;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,kCAAkC,CAAC;IACzC,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,mEAAmE;IACnE,UAAU,EAAE,QAAQ,GAAG,QAAQ,GAAG,IAAI,CAAC;IACvC,2EAA2E;IAC3E,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;IAC9B,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC7C;AAQD,MAAM,WAAW,wBAAwB;IACxC;;;;;;OAMG;IACH,SAAS,CAAC,EAAE,QAAQ,GAAG,QAAQ,GAAG,IAAI,CAAC;IACvC,uEAAuE;IACvE,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B;;;;;;;;;;;;;;;;OAgBG;IACH,aAAa,CAAC,EAAE,YAAY,GAAG,OAAO,GAAG,SAAS,CAAC;IACnD;;;;;;OAMG;IACH,iBAAiB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAClC,GAAG,CAAC,EAAE,MAAM,CAAC;CACb;AAwKD,wBAAgB,+BAA+B,CAC9C,WAAW,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,EACtC,kBAAkB,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,EAC/C,OAAO,CAAC,EAAE,wBAAwB,GAChC,0BAA0B,CA8E5B"}

package/dist/lib/request/rate-limit-decision.js

@@ -232,6 +232,34 @@ const BLOCKER_DESCRIPTIONS = new Map([
         description,
     ]),
 ]);
+// Retry-budget exhaustion can end immediately after an upstream attempt,
+// before selection gets another pass to translate that attempt into a live
+// cooldown or breaker. Keep these final verdicts operator-facing without
+// treating them as live blockers: a concurrently opened circuit or cooldown
+// must still take precedence in buildPinnedUnavailableErrorBody below.
+const ATTEMPT_VERDICT_DESCRIPTIONS = new Map([
+    [
+        "auth-failure",
+        {
+            parenthetical: "authentication failure",
+            deadlineNoun: "available-again",
+        },
+    ],
+    [
+        "network-error",
+        {
+            parenthetical: "upstream network error",
+            deadlineNoun: "available-again",
+        },
+    ],
+    [
+        "server-error",
+        {
+            parenthetical: "upstream server error",
+            deadlineNoun: "available-again",
+        },
+    ],
+]);
 /** Whether the sentence has real wording for this token, or must echo it raw. */
 function isDescribedBlocker(skipReason) {
     return BLOCKER_DESCRIPTIONS.has(skipReason);
@@ -252,7 +280,8 @@ function describePinnedBlocker(skipReason, rateLimitBoundsRecovery) {
             deadline: DEADLINE_SENTENCES["available-again"],
         };
     }
-    const described = BLOCKER_DESCRIPTIONS.get(skipReason);
+    const described = BLOCKER_DESCRIPTIONS.get(skipReason) ??
+        ATTEMPT_VERDICT_DESCRIPTIONS.get(skipReason);
     if (described === undefined) {
         // Permanent blockers never reach the deadline clause (the call site
         // suppresses their reset time), and future or internal tokens stay
@@ -311,9 +340,9 @@ export function buildPinnedUnavailableErrorBody(pinnedIndex, accountSkipReasons,
     // recorded verdict that names a real blocker class wins, because the
     // selection-only classes ("missing", "policy-blocked") have no
     // runtime-state equivalent and would be lost. Otherwise the re-read runtime
-    // state is the truth and the recorded token is the retry loop's own
-    // bookkeeping — the dominant path in issue #675, where a 503/429 on the pin
-    // re-enters selection and records "already-attempted" over the real class.
+    // state is the truth and the recorded token is only the earlier attempt
+    // verdict. This also preserves compatibility with legacy callers that may
+    // still supply an internal bookkeeping token.
     const describedSkipReason = skipReason !== null && isDescribedBlocker(skipReason)
         ? skipReason
         : context?.currentSkipReason ?? skipReason;

package/dist/lib/request/response-metadata.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"response-metadata.d.ts","sourceRoot":"","sources":["../../../lib/request/response-metadata.ts"],"names":[],"mappings":"AASA,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,MAAM,GAAG,IAAI,CAyBrE;AAED,wBAAgB,6BAA6B,CAC5C,OAAO,EAAE,OAAO,GACd,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CA0BxB"}
+{"version":3,"file":"response-metadata.d.ts","sourceRoot":"","sources":["../../../lib/request/response-metadata.ts"],"names":[],"mappings":"AASA,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,OAAO,GAAG,MAAM,GAAG,IAAI,CAyBrE;AAED,wBAAgB,6BAA6B,CAC5C,OAAO,EAAE,OAAO,GACd,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CA2BxB"}

package/dist/lib/request/response-metadata.js

@@ -45,6 +45,7 @@ export function sanitizeResponseHeadersForLog(headers) {
         "x-codex-secondary-window-minutes",
         "x-codex-secondary-reset-at",
         "x-codex-secondary-reset-after-seconds",
+        "x-codex-context-budget-percent",
         "retry-after",
         "x-ratelimit-reset",
         "x-ratelimit-reset-requests",

package/dist/lib/request/stream-failover-runtime.d.ts

@@ -11,5 +11,7 @@ export declare function forwardStreamingResponse(upstream: Response, res: Server
  * chunk and is called before the write, so a slow observer delays the
  * forward. Implementations are expected to swallow their own errors.
  */
-onChunk?: (chunk: Uint8Array) => void): Promise<boolean>;
+onChunk?: (chunk: Uint8Array) => void, 
+/** Additional response headers, applied after (so they can override) the forwarded upstream ones. */
+extraHeaders?: Record<string, string>): Promise<boolean>;
 //# sourceMappingURL=stream-failover-runtime.d.ts.map

package/dist/lib/request/stream-failover-runtime.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stream-failover-runtime.d.ts","sourceRoot":"","sources":["../../../lib/request/stream-failover-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAChD,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,qCAAqC,CAAC;AAEtF,eAAO,MAAM,kBAAkB,aAW7B,CAAC;AAWH,wBAAgB,wBAAwB,CAAC,eAAe,EAAE,OAAO,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAUzF;AAED,wBAAsB,WAAW,CAAC,CAAC,EAClC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,EACnB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,IAAI,EACrB,OAAO,EAAE,MAAM,GACb,OAAO,CAAC,CAAC,CAAC,CAmBZ;AAED,wBAAsB,aAAa,CAClC,QAAQ,EAAE,QAAQ,EAClB,SAAS,EAAE,MAAM,EACjB,QAAQ,SAAc,GACpB,OAAO,CAAC,MAAM,CAAC,CAuDjB;AAoBD,wBAAsB,wBAAwB,CAC7C,QAAQ,EAAE,QAAQ,EAClB,GAAG,EAAE,cAAc,EACnB,MAAM,EAAE,0BAA0B,EAClC,aAAa,EAAE,MAAM,IAAI,EACzB,oBAAoB,EAAE,MAAM;AAC5B;;;;;GAKG;AACH,OAAO,CAAC,EAAE,CAAC,KAAK,EAAE,UAAU,KAAK,IAAI,GACnC,OAAO,CAAC,OAAO,CAAC,CAsDlB"}
+{"version":3,"file":"stream-failover-runtime.d.ts","sourceRoot":"","sources":["../../../lib/request/stream-failover-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,WAAW,CAAC;AAChD,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,qCAAqC,CAAC;AAEtF,eAAO,MAAM,kBAAkB,aAW7B,CAAC;AAWH,wBAAgB,wBAAwB,CAAC,eAAe,EAAE,OAAO,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAUzF;AAED,wBAAsB,WAAW,CAAC,CAAC,EAClC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,EACnB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,IAAI,EACrB,OAAO,EAAE,MAAM,GACb,OAAO,CAAC,CAAC,CAAC,CAmBZ;AAED,wBAAsB,aAAa,CAClC,QAAQ,EAAE,QAAQ,EAClB,SAAS,EAAE,MAAM,EACjB,QAAQ,SAAc,GACpB,OAAO,CAAC,MAAM,CAAC,CAuDjB;AAoBD,wBAAsB,wBAAwB,CAC7C,QAAQ,EAAE,QAAQ,EAClB,GAAG,EAAE,cAAc,EACnB,MAAM,EAAE,0BAA0B,EAClC,aAAa,EAAE,MAAM,IAAI,EACzB,oBAAoB,EAAE,MAAM;AAC5B;;;;;GAKG;AACH,OAAO,CAAC,EAAE,CAAC,KAAK,EAAE,UAAU,KAAK,IAAI;AACrC,qGAAqG;AACrG,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GACnC,OAAO,CAAC,OAAO,CAAC,CAsDlB"}

package/dist/lib/request/stream-failover-runtime.js

@@ -135,9 +135,14 @@ export async function forwardStreamingResponse(upstream, res, status, onStreamEr
  * chunk and is called before the write, so a slow observer delays the
  * forward. Implementations are expected to swallow their own errors.
  */
-onChunk) {
+onChunk, 
+/** Additional response headers, applied after (so they can override) the forwarded upstream ones. */
+extraHeaders) {
     status.streamsStarted += 1;
-    res.writeHead(upstream.status, responseHeadersForClient(upstream.headers));
+    res.writeHead(upstream.status, {
+        ...responseHeadersForClient(upstream.headers),
+        ...extraHeaders,
+    });
     if (!upstream.body) {
         res.end();
         return true;

package/dist/lib/runtime/context-budget-settings.d.ts

@@ -0,0 +1,13 @@
+export declare function applyContextBudgetGuardSettingsFromConfig(pluginConfig: ReturnType<typeof import("../config.js").loadPluginConfig>, deps: {
+    configure: (options: {
+        enabled: boolean;
+        softPercent: number;
+        hardPercent: number;
+        modelWindowOverrides: Record<string, number>;
+    }) => void;
+    getContextBudgetGuardEnabled: (config: ReturnType<typeof import("../config.js").loadPluginConfig>) => boolean;
+    getContextBudgetGuardSoftPercent: (config: ReturnType<typeof import("../config.js").loadPluginConfig>) => number;
+    getContextBudgetGuardHardPercent: (config: ReturnType<typeof import("../config.js").loadPluginConfig>) => number;
+    getContextBudgetGuardModelWindowOverrides: (config: ReturnType<typeof import("../config.js").loadPluginConfig>) => Record<string, number>;
+}): void;
+//# sourceMappingURL=context-budget-settings.d.ts.map

package/dist/lib/runtime/context-budget-settings.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-budget-settings.d.ts","sourceRoot":"","sources":["../../../lib/runtime/context-budget-settings.ts"],"names":[],"mappings":"AAAA,wBAAgB,yCAAyC,CACxD,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,EACxE,IAAI,EAAE;IACL,SAAS,EAAE,CAAC,OAAO,EAAE;QACpB,OAAO,EAAE,OAAO,CAAC;QACjB,WAAW,EAAE,MAAM,CAAC;QACpB,WAAW,EAAE,MAAM,CAAC;QACpB,oBAAoB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;KAC7C,KAAK,IAAI,CAAC;IACX,4BAA4B,EAAE,CAC7B,MAAM,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KAC9D,OAAO,CAAC;IACb,gCAAgC,EAAE,CACjC,MAAM,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KAC9D,MAAM,CAAC;IACZ,gCAAgC,EAAE,CACjC,MAAM,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KAC9D,MAAM,CAAC;IACZ,yCAAyC,EAAE,CAC1C,MAAM,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KAC9D,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC5B,GACC,IAAI,CAQN"}

package/dist/lib/runtime/context-budget-settings.js

@@ -0,0 +1,8 @@
+export function applyContextBudgetGuardSettingsFromConfig(pluginConfig, deps) {
+    deps.configure({
+        enabled: deps.getContextBudgetGuardEnabled(pluginConfig),
+        softPercent: deps.getContextBudgetGuardSoftPercent(pluginConfig),
+        hardPercent: deps.getContextBudgetGuardHardPercent(pluginConfig),
+        modelWindowOverrides: deps.getContextBudgetGuardModelWindowOverrides(pluginConfig),
+    });
+}

package/dist/lib/runtime/loader-setup.d.ts

@@ -5,5 +5,6 @@ export declare function applyLoaderRuntimeSetup(params: {
     ensureSessionAffinity: (pluginConfig: ReturnType<typeof import("../config.js").loadPluginConfig>) => void;
     ensureRefreshGuardian: (pluginConfig: ReturnType<typeof import("../config.js").loadPluginConfig>) => void;
     applyPreemptiveQuotaSettings: (pluginConfig: ReturnType<typeof import("../config.js").loadPluginConfig>) => void;
+    applyContextBudgetGuardSettings: (pluginConfig: ReturnType<typeof import("../config.js").loadPluginConfig>) => void;
 }): void;
 //# sourceMappingURL=loader-setup.d.ts.map

package/dist/lib/runtime/loader-setup.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"loader-setup.d.ts","sourceRoot":"","sources":["../../../lib/runtime/loader-setup.ts"],"names":[],"mappings":"AAAA,wBAAgB,uBAAuB,CAAC,MAAM,EAAE;IAC/C,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,CAAC;IACzE,wBAAwB,EAAE,CACzB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,wBAAwB,EAAE,CACzB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,qBAAqB,EAAE,CACtB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,qBAAqB,EAAE,CACtB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,4BAA4B,EAAE,CAC7B,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;CACV,GAAG,IAAI,CAMP"}
+{"version":3,"file":"loader-setup.d.ts","sourceRoot":"","sources":["../../../lib/runtime/loader-setup.ts"],"names":[],"mappings":"AAAA,wBAAgB,uBAAuB,CAAC,MAAM,EAAE;IAC/C,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,CAAC;IACzE,wBAAwB,EAAE,CACzB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,wBAAwB,EAAE,CACzB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,qBAAqB,EAAE,CACtB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,qBAAqB,EAAE,CACtB,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,4BAA4B,EAAE,CAC7B,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;IACV,+BAA+B,EAAE,CAChC,YAAY,EAAE,UAAU,CAAC,cAAc,cAAc,EAAE,gBAAgB,CAAC,KACpE,IAAI,CAAC;CACV,GAAG,IAAI,CAOP"}

package/dist/lib/runtime/loader-setup.js

@@ -4,4 +4,5 @@ export function applyLoaderRuntimeSetup(params) {
     params.ensureSessionAffinity(params.pluginConfig);
     params.ensureRefreshGuardian(params.pluginConfig);
     params.applyPreemptiveQuotaSettings(params.pluginConfig);
+    params.applyContextBudgetGuardSettings(params.pluginConfig);
 }

package/dist/lib/runtime/rotation-account-selection.d.ts

@@ -34,5 +34,28 @@ export declare function chooseAccount(params: {
     stickyBoostByAccount?: Record<number, number>;
     pidOffsetEnabled?: boolean;
     schedulingStrategy?: "hybrid" | "sequential";
+    /**
+     * Admit a pinned account that is only blocked by its own cooldown.
+     *
+     * Set by the proxy for the RETRY passes of one pinned request, never for
+     * the first pass. A cooldown exists to keep OTHER requests off a flaky
+     * account; a retry budget exists to give THIS request another attempt.
+     * Conflating them meant every transient branch cooled the pin down before
+     * the next selection pass and the budget could never be spent — a pinned
+     * request got exactly one upstream attempt under shipped defaults.
+     *
+     * Only the cooldown is waived. Rate limits, an open circuit, a disabled
+     * account, workspace and policy blocks all still reject the pin, and the
+     * cooldown itself stays on the account for everyone else and for the 503's
+     * `retry_after_ms`.
+     *
+     * One ordering note: `getAccountRuntimeSkipReason` reports the FIRST
+     * blocker it finds and checks the cooldown before the circuit, so an
+     * account that is both cooling down and circuit-open reads as a cooldown
+     * and is admitted here. That is safe: `consumeTokenWithReason` takes the
+     * breaker's admission slot immediately after selection and rejects it with
+     * `circuit-open`, which is also the more accurate skip reason to report.
+     */
+    allowPinnedCooldown?: boolean;
 }): ManagedAccount | null;
 //# sourceMappingURL=rotation-account-selection.d.ts.map

package/dist/lib/runtime/rotation-account-selection.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"rotation-account-selection.d.ts","sourceRoot":"","sources":["../../../lib/runtime/rotation-account-selection.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AACrE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACvD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACzE,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAEnE;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,aAAa,CAAC,MAAM,EAAE;IACrC,cAAc,EAAE,cAAc,CAAC;IAC/B,oBAAoB,EAAE,oBAAoB,GAAG,IAAI,CAAC;IAClD,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,MAAM,EAAE,WAAW,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,gBAAgB,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IACtC,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,qBAAqB,GAAG,IAAI,CAAC;IACrC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,WAAW,CAAC,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,kBAAkB,CAAC,EAAE,QAAQ,GAAG,YAAY,CAAC;CAC7C,GAAG,cAAc,GAAG,IAAI,CAwJxB"}
+{"version":3,"file":"rotation-account-selection.d.ts","sourceRoot":"","sources":["../../../lib/runtime/rotation-account-selection.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AACrE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACvD,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACzE,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAWnE;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,aAAa,CAAC,MAAM,EAAE;IACrC,cAAc,EAAE,cAAc,CAAC;IAC/B,oBAAoB,EAAE,oBAAoB,GAAG,IAAI,CAAC;IAClD,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,MAAM,EAAE,WAAW,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,gBAAgB,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;IACtC,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,qBAAqB,GAAG,IAAI,CAAC;IACrC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,WAAW,CAAC,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,oBAAoB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC9C,gBAAgB,CAAC,EAAE,OAAO,CAAC;IAC3B,kBAAkB,CAAC,EAAE,QAAQ,GAAG,YAAY,CAAC;IAC7C;;;;;;;;;;;;;;;;;;;;;OAqBG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC9B,GAAG,cAAc,GAAG,IAAI,CA0JxB"}

package/dist/lib/runtime/rotation-account-selection.js

@@ -1,3 +1,11 @@
+/**
+ * Whether a runtime skip reason is a cooldown, the one blocker a pinned
+ * retry may waive. `getManagedAccountRuntimeSkipReason` emits either the
+ * bare token or `cooling-down:<reason>`.
+ */
+function isCooldownReason(reason) {
+    return reason === "cooling-down" || reason.startsWith("cooling-down:");
+}
 /**
  * `chooseAccount` is a SYNC selector that internally advances the rotation
  * cursor (the session-affinity-preferred branch and the round-robin fallback
@@ -17,15 +25,16 @@
  * `test/runtime-rotation-proxy.test.ts`.
  */
 export function chooseAccount(params) {
-    const { accountManager, sessionAffinityStore, sessionKey, family, model, attemptedIndexes, now, policy, pinnedIndex, skipReasons, stickyBoostByAccount, pidOffsetEnabled, schedulingStrategy, } = params;
+    const { accountManager, sessionAffinityStore, sessionKey, family, model, attemptedIndexes, now, policy, pinnedIndex, skipReasons, stickyBoostByAccount, pidOffsetEnabled, schedulingStrategy, allowPinnedCooldown, } = params;
     // Manual pin (from `codex-multi-auth switch <n>`) overrides every other
     // selection signal. We do NOT call markSwitched here — the proxy must not
     // clobber the pin set by the CLI. See issue #474.
     if (typeof pinnedIndex === "number") {
-        if (attemptedIndexes.has(pinnedIndex)) {
-            skipReasons?.set(pinnedIndex, "already-attempted");
-            return null;
-        }
+        // A pin has no alternative account to rotate to. Re-evaluate its live
+        // blockers on every pass so a healthy account can use the caller's bounded
+        // retry budget; attemptedIndexes remains authoritative for unpinned pool
+        // selection below. The proxy also carries an absolute loop guard for retry
+        // branches that do not advance their transient-attempt counter.
         if (pinnedIndex < 0 || pinnedIndex >= accountManager.getAccountCount()) {
             skipReasons?.set(pinnedIndex, "missing");
             return null;
@@ -40,7 +49,7 @@ export function chooseAccount(params) {
             return null;
         }
         const reason = accountManager.getAccountRuntimeSkipReason(pinnedIndex, family, model);
-        if (reason) {
+        if (reason && !(allowPinnedCooldown === true && isCooldownReason(reason))) {
             skipReasons?.set(pinnedIndex, reason);
             return null;
         }

package/dist/lib/runtime/rotation-proxy-state.d.ts

@@ -1,4 +1,5 @@
 import { AccountManager } from "../accounts.js";
+import type { ContextBudgetGuard } from "../context-budget-guard.js";
 import type { PreemptiveQuotaScheduler } from "../preemptive-quota-scheduler.js";
 import type { RuntimeRotationProxyStatus } from "./rotation-server-types.js";
 import type { SessionAffinityStore } from "../session-affinity.js";
@@ -29,6 +30,7 @@ export interface RotationProxyStateInit {
     maxRequestBodyBytes: number;
     quotaRemainingPercentThreshold: number;
     preemptiveQuotaScheduler: PreemptiveQuotaScheduler;
+    contextBudgetGuard: ContextBudgetGuard;
     sessionAffinityStore: SessionAffinityStore | null;
     lastObservedAffinityGeneration: number;
     /**

package/dist/lib/runtime/rotation-proxy-state.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"rotation-proxy-state.d.ts","sourceRoot":"","sources":["../../../lib/runtime/rotation-proxy-state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,kCAAkC,CAAC;AAKjF,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,4BAA4B,CAAC;AAC7E,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAEnE;;;;;;GAMG;AACH,MAAM,WAAW,sBAAsB;IACtC,oBAAoB,EAAE,cAAc,CAAC;IACrC,gBAAgB,EAAE,SAAS,GAAG,QAAQ,CAAC;IACvC,kBAAkB,EAAE,QAAQ,GAAG,YAAY,CAAC;IAC5C,SAAS,EAAE,OAAO,KAAK,CAAC;IACxB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,GAAG,EAAE,MAAM,MAAM,CAAC;IAClB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,sBAAsB,EAAE,MAAM,CAAC;IAC/B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,2BAA2B,EAAE,MAAM,CAAC;IACpC,qBAAqB,EAAE,MAAM,CAAC;IAC9B,gBAAgB,EAAE,OAAO,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC;IAC7B,yBAAyB,EAAE,MAAM,CAAC;IAClC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,8BAA8B,EAAE,MAAM,CAAC;IACvC,wBAAwB,EAAE,wBAAwB,CAAC;IACnD,oBAAoB,EAAE,oBAAoB,GAAG,IAAI,CAAC;IAClD,8BAA8B,EAAE,MAAM,CAAC;IACvC;;;;;OAKG;IACH,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAC;CAClC;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,kBAAmB,SAAQ,sBAAsB;IACjE,QAAQ,CAAC,oBAAoB,EAAE,GAAG,CAAC,cAAc,CAAC,CAAC;IACnD,QAAQ,CAAC,MAAM,EAAE,0BAA0B,CAAC;IAC5C,QAAQ,CAAC,mBAAmB,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,CAAC;IACzD,sBAAsB,EAAE,MAAM,GAAG,IAAI,CAAC;IACtC,kBAAkB,EAAE,MAAM,CAAC;IAC3B,yBAAyB,EAAE,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;IACjE,wBAAwB,EAAE,MAAM,CAAC;CACjC;AAID,gBAAgB;AAChB,wBAAgB,wBAAwB,CACvC,IAAI,EAAE,sBAAsB,GAC1B,kBAAkB,CAuBpB;AAED,gBAAgB;AAChB,wBAAsB,wBAAwB,CAC7C,KAAK,EAAE,kBAAkB,GACvB,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC,CAkDhC"}
+{"version":3,"file":"rotation-proxy-state.d.ts","sourceRoot":"","sources":["../../../lib/runtime/rotation-proxy-state.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAChD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AACrE,OAAO,KAAK,EAAE,wBAAwB,EAAE,MAAM,kCAAkC,CAAC;AAKjF,OAAO,KAAK,EAAE,0BAA0B,EAAE,MAAM,4BAA4B,CAAC;AAC7E,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAEnE;;;;;;GAMG;AACH,MAAM,WAAW,sBAAsB;IACtC,oBAAoB,EAAE,cAAc,CAAC;IACrC,gBAAgB,EAAE,SAAS,GAAG,QAAQ,CAAC;IACvC,kBAAkB,EAAE,QAAQ,GAAG,YAAY,CAAC;IAC5C,SAAS,EAAE,OAAO,KAAK,CAAC;IACxB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,GAAG,EAAE,MAAM,MAAM,CAAC;IAClB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,sBAAsB,EAAE,MAAM,CAAC;IAC/B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,2BAA2B,EAAE,MAAM,CAAC;IACpC,qBAAqB,EAAE,MAAM,CAAC;IAC9B,gBAAgB,EAAE,OAAO,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,oBAAoB,EAAE,MAAM,CAAC;IAC7B,yBAAyB,EAAE,MAAM,CAAC;IAClC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,8BAA8B,EAAE,MAAM,CAAC;IACvC,wBAAwB,EAAE,wBAAwB,CAAC;IACnD,kBAAkB,EAAE,kBAAkB,CAAC;IACvC,oBAAoB,EAAE,oBAAoB,GAAG,IAAI,CAAC;IAClD,8BAA8B,EAAE,MAAM,CAAC;IACvC;;;;;OAKG;IACH,kBAAkB,EAAE,MAAM,GAAG,IAAI,CAAC;CAClC;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,kBAAmB,SAAQ,sBAAsB;IACjE,QAAQ,CAAC,oBAAoB,EAAE,GAAG,CAAC,cAAc,CAAC,CAAC;IACnD,QAAQ,CAAC,MAAM,EAAE,0BAA0B,CAAC;IAC5C,QAAQ,CAAC,mBAAmB,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC,CAAC;IACzD,sBAAsB,EAAE,MAAM,GAAG,IAAI,CAAC;IACtC,kBAAkB,EAAE,MAAM,CAAC;IAC3B,yBAAyB,EAAE,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;IACjE,wBAAwB,EAAE,MAAM,CAAC;CACjC;AAID,gBAAgB;AAChB,wBAAgB,wBAAwB,CACvC,IAAI,EAAE,sBAAsB,GAC1B,kBAAkB,CAuBpB;AAED,gBAAgB;AAChB,wBAAsB,wBAAwB,CAC7C,KAAK,EAAE,kBAAkB,GACvB,OAAO,CAAC,cAAc,GAAG,IAAI,CAAC,CAkDhC"}

package/dist/lib/runtime/rotation-server-types.d.ts

@@ -58,6 +58,17 @@ export interface RequestContext {
     family: ModelFamily;
     stream: boolean;
     sessionKey: string | null;
+    /**
+     * `sessionKey` minus the `previous_response_id` fallback.
+     *
+     * `previous_response_id` changes on every turn, which is fine for session
+     * affinity (a fresh key just means "no pin yet") but wrong for anything
+     * that has to ACCUMULATE across turns: a per-turn key never finds what the
+     * previous turn stored, and leaves one dead map entry behind per request.
+     * Consumers that track state over a conversation use this instead, and
+     * no-op when it is null.
+     */
+    stableSessionKey: string | null;
 }
 export type ExhaustionReason = "rate-limit" | "server-error" | "network-error" | "auth-failure" | "budget" | "deactivated" | "no-account";
 export type RuntimeProxyHttpError = Error & {