@oh-my-pi/pi-ai 18.8.3 → 18.8.5

package/CHANGELOG.md

@@ -2,6 +2,47 @@
 
 ## [Unreleased]
 
+## [18.8.5] - 2026-10-08
+
+### Added
+
+- `oauth.refresh(id, signal, { reason: "auth-recovery" })` forwards provider-401 recovery intent to a delegated (auth broker) refresh, and `AuthStorageOptions.refreshOAuthCredentialMints` marks a `refreshOAuthCredential` hook that exchanges tokens itself so its tokens are reused for auth recovery ([#14752](https://github.com/can1357/oh-my-pi/pull/14752) by [@will-bogusz](https://github.com/will-bogusz))
+- `SessionsApi.inherit` accepts an optional filter, called with each provider and whether the source's affinity is an explicit user pin, to copy only the affinities it accepts ([#14749](https://github.com/can1357/oh-my-pi/pull/14749) by [@will-bogusz](https://github.com/will-bogusz))
+
+### Changed
+
+- `AuthApiKeyOptions.accountIds` also matches the login email, or else the project id, of credentials that carry no account id (see `oauthAccountKey`), so Antigravity requests prefer accounts that serve the requested model ([#14924](https://github.com/can1357/oh-my-pi/issues/14924)).
+
+### Fixed
+
+- Fixed accounts sitting exactly at their `reservePct` (e.g. 70% used with a 30% reserve) still being picked and reported healthy instead of being held in reserve ([#14765](https://github.com/can1357/oh-my-pi/pull/14765) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed `oauth.accessById` with `forceRefresh` returning the stored token unchanged while it was still valid; it now re-mints that one account (through the auth broker when configured) and returns that account's token even if another row is removed meanwhile ([#14752](https://github.com/can1357/oh-my-pi/pull/14752) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed `omp -p` and other short-lived auth-broker clients missing from `omp usage clients`: usage still waiting for the 10-second report batch is now sent to the broker before the process exits ([#14899](https://github.com/can1357/oh-my-pi/pull/14899) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed Cursor web fetches that were cut off by a dropped connection disappearing from resumed and rebuilt sessions; they now show as interrupted ([#14819](https://github.com/can1357/oh-my-pi/pull/14819) by [@jchanghong023](https://github.com/jchanghong023))
+- Anthropic hosted web search can honor custom providers' OAuth-style request shaping and configured headers consistently with conversations ([#14919](https://github.com/can1357/oh-my-pi/pull/14919) by [@farnoy](https://github.com/farnoy))
+
+## [18.8.4] - 2026-10-08
+
+### Breaking Changes
+
+- `AuthBrokerClient.notifyUsageStale` and `UsageLedgerStore.invalidateUsageCache` (including `RemoteAuthCredentialStore.invalidateUsageCache`) now take an optional leading `provider` argument: the signature is `(provider?: string, signal?: AbortSignal)` instead of `(signal?: AbortSignal)` ([#14761](https://github.com/can1357/oh-my-pi/pull/14761) by [@will-bogusz](https://github.com/will-bogusz))
+
+### Added
+
+- Auth gateway route option `excludeProviders` leaves those providers' accounts out of `/v1/usage` and `/v1/credentials/check` ([#14755](https://github.com/can1357/oh-my-pi/pull/14755) by [@will-bogusz](https://github.com/will-bogusz))
+
+### Fixed
+
+- Fixed a single-provider usage refresh from an auth broker client (`omp usage invalidate --provider`, Codex auto-redeem, a saved-reset redeem) wiping the broker's cached usage for every provider, so other accounts showed no usage on every client until their next successful probe ([#14761](https://github.com/can1357/oh-my-pi/pull/14761) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed Anthropic sessions missing the prompt cache a second time after the API dropped thinking blocks from a changed conversation ([#14748](https://github.com/can1357/oh-my-pi/pull/14748) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed Cursor logins never renewing: every refresh was rejected with "Invalid User API Key", so logins lapsed about 60 days after sign-in; a session Cursor has ended is now disabled with a re-login hint ([#14753](https://github.com/can1357/oh-my-pi/pull/14753) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed the auth broker logging caller-supplied `X-Forwarded-For` / `X-Real-IP` values and unknown request paths; peers come from the socket unless `trustProxyHeaders` is set ([#14762](https://github.com/can1357/oh-my-pi/pull/14762) by [@will-bogusz](https://github.com/will-bogusz))
+- Fixed false thinking-loop detections on Gemini, DeepSeek, and Grok when the model drafts or answers with repetitive code or markup (VRML, SVG, JSON); such output is no longer discarded and retried.
+- Fixed Cursor provider leaking conversation checkpoints, blob stores and abort listeners after requests end ([#14669](https://github.com/can1357/oh-my-pi/pull/14669) by [@H4vC](https://github.com/H4vC))
+- Fixed session-affinity pins growing without bound in long-lived gateways (now capped at 256 sessions per provider) ([#14672](https://github.com/can1357/oh-my-pi/pull/14672) by [@H4vC](https://github.com/H4vC))
+- Fixed completed Cursor turns failing with "Cursor stream ended before turnEnded" when the connection closed after the answer had fully arrived ([#14851](https://github.com/can1357/oh-my-pi/pull/14851) by [@kyle-elliott-asymptote](https://github.com/kyle-elliott-asymptote)).
+- Fixed Cursor provider errors that Cursor marks as not retryable being retried until the retry budget ran out ([#14851](https://github.com/can1357/oh-my-pi/pull/14851) by [@kyle-elliott-asymptote](https://github.com/kyle-elliott-asymptote)).
+
 ## [18.8.3] - 2026-10-07
 
 ### Added
@@ -2307,67 +2348,4 @@
 - Fixed `validateToolArguments` silently accepting JSON-encoded array strings (e.g. `'["a","b"]'`) against `union(string, array<string>)` schemas — providers that double-serialize tool-call arguments (Z.AI / GLM) caused tools like `search` to receive the literal `["a","b"]` as a single path, producing zero matches (single element) or glob parse errors (multi-element). A new pre-validation pass parses JSON-array-shaped strings when the schema explicitly accepts both shapes. ([#1788](https://github.com/can1357/oh-my-pi/issues/1788))
 - Fixed Anthropic thinking summaries that arrive wrapped in literal `<thinking>` tags so advisor/raw transcript dumps do not render nested thinking tags ([#2695](https://github.com/can1357/oh-my-pi/issues/2695)).
 
-## [16.0.0] - 2026-06-15
-
-### Breaking Changes
-
-- Renamed the public dialect entrypoint from `@oh-my-pi/pi-ai/grammar` to `@oh-my-pi/pi-ai/dialect`.
-- Renamed grammar dialect identifiers from `ToolCallSyntax` to `Dialect`, renamed the `Grammar` interface to `DialectDefinition`, and renamed `Grammar.syntax` to `DialectDefinition.dialect`.
-- Added `DialectDefinition.renderThinking` and `DialectDefinition.renderTranscript` so dialect implementations serialize complete native chat transcripts, not just tool call/result blocks.
-
-### Added
-
-- Added `renderTranscript` method to dialect definitions for serializing complete native chat transcripts
-- Added `renderThinking` method to dialect definitions for rendering thinking/reasoning blocks
-- Added support for 11 dialect implementations: Anthropic, DeepSeek, Gemini, Gemma, GLM, Harmony, Hermes, Kimi, Pi-native, Qwen3, and XML
-- Added `createInbandScanner` factory function to instantiate dialect-specific scanners
-- Added `getDialectDefinition` function to retrieve dialect implementations by name
-- Added `renderToolCatalog` and `renderInbandToolPrompt` functions for tool catalog rendering
-- Added `renderToolInventory` function to generate human-readable per-tool documentation with examples
-- Added `renderToolExamples` function to render tool usage examples in the model's native dialect
-- Added `encodeInbandToolHistory` function to encode tool call history in dialect-specific format
-- Added `wrapInbandToolStream` function to process streaming responses with in-band tool call parsing
-- Added `ThinkingInbandScanner` for parsing thinking/reasoning blocks across dialects
-- Added `OwnedStream` class for managing dialect-aware streaming with tool call events
-- Added in-band thinking channels to every dialect that was missing one: `gemini` (a ` ```thinking ` fence mirroring ` ```tool_code `), `gemma` (its native `<|channel>thought…<channel|>` reasoning channel), `kimi` (`<think>…</think>`), and `pi` (`<thinking>…</thinking>`). Each scanner now parses reasoning into thinking events instead of leaking chain-of-thought into the visible reply, and every dialect's `renderThinking` is a real channel that round-trips back through its scanner (no passthrough renderers).
-
-### Changed
-
-- Moved public dialect entrypoint from `@oh-my-pi/pi-ai/grammar` to `@oh-my-pi/pi-ai/dialect` in package exports
-- Updated internal imports in `stream-markup-healing.ts` to use new dialect module path
-- Changed `renderToolInventory` to demote a tool description's own markdown headers by one level when it contains a top-level `# ` header, so they nest under the wrapping `# Tool: <name>` heading instead of reading as sibling sections. Descriptions that already start at `##` and headers inside fenced code blocks are left untouched.
-
-### Fixed
-
-- Fixed Gemini, Gemma, Kimi, and Pi in-band scanners to respect `parseThinking: false`, leaving private reasoning markers in visible text when parsing is disabled
-- Fixed thinking-channel parsing for streaming Gemini, Gemma, Kimi, and Pi outputs so split or partial `<thinking>` blocks no longer leak into visible replies
-- Fixed in-band thinking finalization and Kimi stream-healing interactions so leaked `<think>` blocks are preserved when structured tool calls are present, not duplicated when explicit reasoning is present, and closed on stream flush.
-
-### Removed
-
-- Removed `src/grammar/factory.ts` (replaced by `src/dialect/factory.ts`)
-- Removed `src/grammar/rendering.ts` (functionality moved to `src/dialect/rendering.ts`)
-- Removed `src/grammar/xml.ts` (replaced by `src/dialect/xml.ts`)
-
-## [15.13.3] - 2026-06-15
-
-### Added
-
-- Added the `gemini` in-band tool-call syntax with Python-style `tool_code` blocks and `default_api` invocations
-- Added the `gemma` token-delimited in-band tool-call syntax using `<|tool_call>` and `<|tool_response>` blocks
-- Added `gemini` and `gemma` to owned stream tool-result token detection so their tool responses are recognized
-- Fixed truncated Gemini and Gemma tool blocks from being emitted as plain text during streaming
-- Added the Azure OpenAI provider definition (`azure`) to the registry; `AZURE_OPENAI_API_KEY` resolves as its env-var API key via the catalog provider table.
-
-### Changed
-
-- Gemini tool-call examples now render without the `default_api.` namespace prefix, keeping `<example>` blocks concise. The live wire format still uses `default_api.` per the Gemini grammar.
-
-### Fixed
-
-- Fixed duplicate tool call projections by deduplicating provider-native `toolCall` events against in-band `tool_code` calls and keeping only the first real channel
-- Dropped nameless native `toolCall` events so they no longer appear as surfaced tool calls in owned-mode streams
-- Fixed truncated Gemini and Gemma tool blocks from being emitted as plain text during streaming
-- Fixed Gemini/Gemma in-band tool-call parsing around Python comments, raw/unicode string literals, and Gemma close-token text inside string values.
-
-Older entries are archived in [packages/ai/CHANGELOG.md@a20cc0d04b64](https://github.com/can1357/oh-my-pi/blob/a20cc0d04b64f4a68fcc559cbd258743d984c50e/packages/ai/CHANGELOG.md).
+Older entries are archived in [packages/ai/CHANGELOG.md@d22e34333695](https://github.com/can1357/oh-my-pi/blob/d22e34333695d3dd9de7741ff45903681dba4606/packages/ai/CHANGELOG.md).

package/dist/types/auth/affinity.d.ts

@@ -79,14 +79,15 @@ export declare class SessionAffinity implements SessionsApi {
         restoredAtMs?: number;
     }): boolean;
     /**
-     * Copy every stored credential affinity from one live session to another.
+     * Copy stored credential affinities from one live session to another.
      *
      * The target receives its own sticky entries, so request resolution, usage
      * blocking, credential rotation, metadata, and persisted pins all continue
      * through the target session id without retaining a live dependency on the
-     * source session.
+     * source session. `include` limits the copy to the providers it accepts,
+     * given whether the source's affinity is an explicit user pin.
      */
-    inherit(sourceSessionId: string, targetSessionId: string): number;
+    inherit(sourceSessionId: string, targetSessionId: string, include?: (provider: string, explicit: boolean) => boolean): number;
     /**
      * Release a session's sticky credential so its next `KeyCascade.get` call
      * re-runs native pool ranking. This never blocks or penalizes the released

package/dist/types/auth/oauth.d.ts

@@ -67,9 +67,10 @@ export declare class OAuthAccounts implements OAuthApi {
      * Resolve one stored OAuth credential by its durable storage row id.
      *
      * Unlike the normal session resolver, this method never ranks, rotates, or
-     * falls back to sibling credentials. A forced refresh re-mints only the
-     * requested row, preserving exact-account affinity for operations whose
-     * provenance and policy boundary are tied to one workspace.
+     * falls back to sibling credentials. A generic forced refresh re-mints only the
+     * requested row; with `refreshReason: "auth-recovery"` it may instead reuse a
+     * still-usable recent mint of that row. Either way exact-account affinity holds
+     * for operations whose provenance and policy boundary are tied to one workspace.
      *
      * Returns `undefined` when the row does not exist for `provider` or an
      * explicit runtime/config API-key override suppresses OAuth.

package/dist/types/auth/refresh.d.ts

@@ -32,6 +32,7 @@ export interface OAuthRefresherDeps {
     pool: CredentialPool;
     policies: AccountPolicies;
     override?: AuthStorageOptions["refreshOAuthCredential"];
+    overrideMints?: AuthStorageOptions["refreshOAuthCredentialMints"];
 }
 /** Single-flighted, lease-guarded OAuth refresh with compare-and-set persistence. */
 export declare class OAuthRefresher {

package/dist/types/auth/select.d.ts

@@ -8,7 +8,7 @@ import { type CredentialPool } from "./pool.js";
 import { type ApiKeySelection, type OAuthSelection } from "./rank.js";
 import { type OAuthRefresher } from "./refresh.js";
 import type { AuthCredentialStore } from "./store.js";
-import type { ApiKeyCredential, AuthApiKeyOptions, AuthCredential, OAuthCredential } from "./types.js";
+import { type ApiKeyCredential, type AuthApiKeyOptions, type AuthCredential, type OAuthCredential } from "./types.js";
 import type { UsageService } from "./usage.js";
 /** Temporary block after a transient OAuth refresh failure. */
 export declare const OAUTH_REFRESH_FAILURE_BACKOFF_MS: number;

package/dist/types/auth/store.d.ts

@@ -122,8 +122,8 @@ export interface UsageLedgerStore {
 }
 /** Broker-delegated OAuth refresh and usage-report operations. */
 export interface CredentialUpstream {
-    /** Optional hook to notify the underlying store that usage report cache is stale. */
-    invalidateUsageCache(signal?: AbortSignal): Promise<void>;
+    /** Optional hook to notify the underlying store that usage reports (one provider's, when given) are stale. */
+    invalidateUsageCache(provider?: string, signal?: AbortSignal): Promise<void>;
     /**
      * Optional store-supplied OAuth refresh. When present, `AuthStorage` uses
      * it before the per-provider local refresh path. `RemoteAuthCredentialStore`

package/dist/types/auth/types.d.ts

@@ -316,6 +316,12 @@ export type AuthStorageOptions = {
      * the actual refresh token never leaves the broker.
      */
     refreshOAuthCredential?: (provider: Provider, credentialId: number, credential: OAuthCredential, signal?: AbortSignal, reason?: OAuthRefreshReason) => Promise<OAuthCredentials>;
+    /**
+     * Set when {@link refreshOAuthCredential} exchanges the refresh token with the
+     * provider itself instead of delegating (e.g. to a broker), so its tokens count
+     * as this process's mints for auth-recovery reuse.
+     */
+    refreshOAuthCredentialMints?: boolean;
     /**
      * Human-readable description of the credential store backing this
      * AuthStorage instance. Surfaced through {@link AuthStorage.keys.describe}
@@ -409,7 +415,7 @@ export interface ModelUsageHealthOptions {
 export type AuthApiKeyOptions = {
     baseUrl?: string;
     modelId?: string;
-    /** Provider account ids known to serve `modelId` from multi-account discovery; OAuth selection prefers them and tries other accounts only as a last resort. */
+    /** {@link oauthAccountKey}s of accounts known to serve `modelId` from multi-account discovery; OAuth selection prefers them and tries other accounts only as a last resort. */
     accountIds?: readonly string[];
     /**
      * Caller's cancel signal. Threaded into any broker-bound OAuth refresh so
@@ -429,6 +435,19 @@ export type AuthApiKeyOptions = {
     /** When false, select as `sessionId` would without recording the choice as that session's sticky credential. */
     recordAffinity?: boolean;
 };
+/**
+ * Identity that multi-account discovery records in `Model.accountAccess` and
+ * that {@link AuthApiKeyOptions.accountIds} selection matches: the provider
+ * account id (Codex `chatgpt_account_id`), else the login email for providers
+ * whose credentials carry no account id (Antigravity), else the project id for
+ * a login whose email lookup failed. Same precedence as the credential store's
+ * identity key.
+ */
+export declare function oauthAccountKey(identity: {
+    accountId?: string;
+    email?: string;
+    projectId?: string;
+}): string | undefined;
 /** Non-secret identity bound to the OAuth credential selected for one request attempt. */
 export interface OAuthRequestIdentity {
     orgId?: string;
@@ -917,7 +936,7 @@ export interface KeysApi {
         modelId?: string;
     }): ApiKeyResolver;
 }
-/** Controls whether a row-id refresh may reuse a token minted by this refresher. */
+/** Controls whether a row-id refresh may reuse a recently minted token. */
 export interface OAuthRefreshByIdOptions {
     /**
      * Return the stored credential when it still holds a fresh access token this
@@ -925,6 +944,12 @@ export interface OAuthRefreshByIdOptions {
      * refresh tokens repeatedly when a provider rejects every valid bearer.
      */
     reuseRecentMint?: boolean;
+    /**
+     * Provider-401 recovery intent. Forwarded to a delegated refresh (an auth
+     * broker), which may then return a token it minted in the last five minutes.
+     * Local reuse still requires {@link reuseRecentMint}.
+     */
+    reason?: OAuthRefreshReason;
 }
 /** OAuth login, access, account identity, and refresh operations. */
 export interface OAuthApi {
@@ -966,9 +991,10 @@ export interface OAuthApi {
      * Resolve one stored OAuth credential by its durable storage row id.
      *
      * Unlike the normal session resolver, this method never ranks, rotates, or
-     * falls back to sibling credentials. A forced refresh re-mints only the
-     * requested row, preserving exact-account affinity for operations whose
-     * provenance and policy boundary are tied to one workspace.
+     * falls back to sibling credentials. A generic forced refresh re-mints only the
+     * requested row; with `refreshReason: "auth-recovery"` it may instead reuse a
+     * still-usable recent mint of that row. Either way exact-account affinity holds
+     * for operations whose provenance and policy boundary are tied to one workspace.
      *
      * Returns `undefined` when the row does not exist for `provider` or an
      * explicit runtime/config API-key override suppresses OAuth.
@@ -1002,8 +1028,9 @@ export interface OAuthApi {
      * Refresh the OAuth credential with the given id through a per-credential
      * single-flight. Concurrent callers for the same row await the same upstream
      * refresh attempt, which is required for providers that rotate refresh tokens
-     * on every successful refresh. Mints unconditionally unless
-     * {@link OAuthRefreshByIdOptions.reuseRecentMint} is set.
+     * on every successful refresh. Mints unless {@link OAuthRefreshByIdOptions.reuseRecentMint}
+     * finds a recent local mint, or a delegated refresh given
+     * {@link OAuthRefreshByIdOptions.reason} `"auth-recovery"` returns its recent mint.
      */
     refresh(id: number, signal?: AbortSignal, options?: OAuthRefreshByIdOptions): Promise<AuthCredentialSnapshotEntry>;
     /**
@@ -1035,14 +1062,15 @@ export interface SessionsApi {
         restoredAtMs?: number;
     }): boolean;
     /**
-     * Copy every stored credential affinity from one live session to another.
+     * Copy stored credential affinities from one live session to another.
      *
      * The target receives its own sticky entries, so request resolution, usage
      * blocking, credential rotation, metadata, and persisted pins all continue
      * through the target session id without retaining a live dependency on the
-     * source session.
+     * source session. `include` limits the copy to the providers it accepts,
+     * given whether the source's affinity is an explicit user pin.
      */
-    inherit(sourceSessionId: string, targetSessionId: string): number;
+    inherit(sourceSessionId: string, targetSessionId: string, include?: (provider: string, explicit: boolean) => boolean): number;
     /**
      * Restrict one session's credentials for `provider` to the OAuth accounts
      * whose identity key (`email:<address>|org:<id>` for org-scoped providers;

package/dist/types/auth-broker/client.d.ts

@@ -85,7 +85,7 @@ export declare class AuthBrokerClient {
     fetchClientUsageSummary(query?: {
         sinceMs?: number;
     }, signal?: AbortSignal): Promise<ClientUsageSummaryResponse>;
-    notifyUsageStale(signal?: AbortSignal): Promise<UsageStaleResponse>;
+    notifyUsageStale(provider?: string, signal?: AbortSignal): Promise<UsageStaleResponse>;
     refreshCredential(id: number, signal?: AbortSignal, reason?: OAuthRefreshReason): Promise<CredentialRefreshResponse>;
     disableCredential(id: number, cause: string, signal?: AbortSignal): Promise<CredentialDisableResponse>;
     /**

package/dist/types/auth-broker/remote-store.d.ts

@@ -122,7 +122,7 @@ export declare class RemoteAuthCredentialStore implements AuthCredentialStore {
     /** Drop all cache rows whose keys start with the supplied prefix. */
     deleteCachePrefix(prefix: string): void;
     cleanExpiredCache(): void;
-    invalidateUsageCache(signal?: AbortSignal): Promise<void>;
+    invalidateUsageCache(provider?: string, signal?: AbortSignal): Promise<void>;
     /**
      * Store-level hook consumed by `AuthStorage` — routes refresh through the
      * broker so the actual refresh token never leaves the broker host. Returns
@@ -149,9 +149,10 @@ export declare class RemoteAuthCredentialStore implements AuthCredentialStore {
     ingestUsageReport(provider: Provider, credential: OAuthCredential, report: UsageReport): boolean;
     /**
      * Fold locally observed request usage into the pending report and schedule
-     * a flush. One `POST /v1/usage/observed` at most per flush interval; on
-     * failure the batch is retained and retried with the next flush. A 404
-     * (pre-endpoint broker) disables reporting for the life of this store.
+     * a flush. One `POST /v1/usage/observed` at most per flush interval, plus a
+     * final one when the process exits; on failure the batch is retained and
+     * retried with the next flush. A 404 (pre-endpoint broker) disables
+     * reporting for the life of this store.
      *
      * `client` overrides the reporting identity — the auth-gateway attributes
      * each request to the originating install/app instead of the gateway host.

package/dist/types/auth-broker/server.d.ts

@@ -17,6 +17,8 @@ export interface AuthBrokerServerOptions {
     bind?: string;
     /** Accept any of these bearer tokens. Empty disables auth (loopback only). */
     bearerTokens: string[];
+    /** Honor forwarded peer headers only when the connecting proxy is trusted. Default false. */
+    trustProxyHeaders?: boolean;
     /** Broker version string surfaced on `/v1/healthz`. */
     version?: string;
     /** Refresh credentials expiring within this window. Default 5 min. */

package/dist/types/auth-gateway/dispatch.d.ts

@@ -17,6 +17,8 @@ export interface AuthGatewayRouteOptions {
     resolveModel: ModelResolver;
     /** Optional supplier for `/v1/models` listing. Returns the full model array. */
     listModels?: () => Iterable<Model<Api>>;
+    /** Providers the host does not serve; `/v1/usage` and `/v1/credentials/check` leave their accounts out. */
+    excludeProviders?: ReadonlySet<string>;
     /** Upstream transport for every provider call; defaults to global `fetch`. Test seam. */
     fetch?: FetchImpl;
 }

package/dist/types/auth-gateway/http.d.ts

@@ -16,8 +16,6 @@ export declare function gatewayResponseHeaders(model: Model<Api>, info: {
     costUsd?: number;
     startedAt?: number;
 }): Record<string, string>;
-/** Use the socket peer unless the gateway explicitly trusts its reverse proxy. */
-export declare function resolvePeer(req: Request, socketAddress: string, trustProxyHeaders?: boolean): string;
 /** Keep configured gateway credentials out of URL and forwarded/logged request fields. */
 export declare function hasMisplacedBearer(req: Request, url: URL, tokens: ReadonlySet<string>): boolean;
 /**

package/dist/types/provider-session-state.d.ts

@@ -18,8 +18,8 @@
  *   siblings are endpoint-scoped and stay: `strictToolsDisabled`
  *   (grammar-too-large 400 for the model's tool schema),
  *   `replayUnsignedThinkingDisabled` / `thinkingReplayDisabled` (the endpoint
- *   is a signing proxy), `prefixDroppedThinkingBlocks` (blocks the API itself
- *   dropped).
+ *   is a signing proxy), `prefixDroppedThinkingBlocks` (blocks a prefix-binding
+ *   400 made the client strip).
  * - **OpenAI Responses** — the `previous_response_id` chain baselines are
  *   account-scoped: a stored response belongs to the account that created it.
  *   Strict-tools / reasoning-effort fallbacks, replay warmup and the chaining

package/dist/types/providers/anthropic.d.ts

@@ -169,6 +169,12 @@ export type AnthropicClientOptionsResult = {
  * omitted there.
  */
 export declare function resolveAnthropicCustomHeadersForBaseUrl(baseUrl: string | undefined): Record<string, string> | undefined;
+/**
+ * Merge headers case-insensitively; later sources win and preserve their key casing.
+ * Undefined sources are ignored. Differently cased duplicates are removed so the
+ * Headers constructor cannot join competing values comma-separated on the wire.
+ */
+export declare function mergeHeaders(...headerSources: (Record<string, string> | undefined)[]): Record<string, string>;
 export type AnthropicUsageLike = {
     cache_creation?: {
         ephemeral_5m_input_tokens?: number | null;

package/dist/types/providers/cursor.d.ts

@@ -169,13 +169,13 @@ export declare function emptyGrepPatternRejection(pattern: string | undefined, g
  * so reparsing unconditionally would erase the arguments of every such block
  * caught open by a truncated stream.
  *
- * Server-owned blocks are also paired here. `connect-scm` and `todo` are
- * stamped {@link kCursorExecResolved} the moment they open, so `agent-loop.ts`
- * synthesizes no placeholder for them and only their `toolCallCompleted` frame
- * pairs a result. A transport that closes before that frame would leave the
- * call unpaired, and `buildSessionContext` strips a dangling call from every
- * rebuilt transcript — the interaction disappears. An interrupted result is
- * emitted instead.
+ * Server-owned blocks are also paired here. `connect-scm`, `todo` and
+ * `web-fetch` are stamped {@link kCursorExecResolved} the moment they open,
+ * so `agent-loop.ts` synthesizes no placeholder for them and only their
+ * `toolCallCompleted` frame pairs a result. A transport that closes before
+ * that frame would leave the call unpaired, and `buildSessionContext` strips
+ * a dangling call from every rebuilt transcript — the interaction disappears.
+ * An interrupted result is emitted instead.
  *
  * MCP blocks are excluded even when resolved: the exec dispatch that marked
  * them owns their result, and `drainInFlightDispatches` awaits it before this

package/dist/types/registry/oauth/cursor.d.ts

@@ -13,7 +13,7 @@ export declare function pollCursorAuth(uuid: string, verifier: string): Promise<
 }>;
 export declare function loginCursor(onAuthUrl: (url: string) => void, onPollStart?: () => void): Promise<OAuthCredentials>;
 export declare function loginCursorHook(callbacks: OAuthController): Promise<OAuthCredentials>;
-export declare function refreshCursorToken(apiKeyOrRefreshToken: string): Promise<OAuthCredentials>;
+export declare function refreshCursorToken(refreshToken: string): Promise<OAuthCredentials>;
 export declare function refreshCursorHook(credentials: OAuthCredentials, signal?: AbortSignal): Promise<OAuthCredentials>;
 /** Request headers that present `accessToken` to cursor.com as its user's web session. */
 export declare function cursorSessionHeaders(userId: string, accessToken: string): Record<string, string>;

package/dist/types/usage.d.ts

@@ -142,6 +142,8 @@ export interface UsageReport {
  * explicit fraction > used/limit > percent-unit used > inverted remaining.
  */
 export declare function resolveUsedFraction(limit: UsageLimit): number | undefined;
+/** Whether remaining quota is at or below the reserve (both 0..1 fractions). */
+export declare function isWithinUsageReserve(remainingFraction: number, reserveFraction: number): boolean;
 /**
  * One recorded usage-limit snapshot: a single limit window of one account at
  * a point in time. The usage cache itself is latest-snapshot-only; history

package/dist/types/utils/anthropic-auth.d.ts

@@ -9,6 +9,11 @@ export declare function resolveAnthropicBaseUrlFromEnv(): string | undefined;
  * Checks if a token is an OAuth token by looking for sk-ant-oat prefix.
  */
 export declare function isOAuthToken(apiKey: string): boolean;
+/** Optional request-shaping overrides for an already-resolved Anthropic credential. */
+export interface AnthropicAuthOptions {
+    /** Override token detection; false explicitly disables OAuth-style request shaping. */
+    isOAuth?: boolean;
+}
 /**
  * Build an {@link AnthropicAuthConfig} from an already-resolved API key.
  *
@@ -17,18 +22,27 @@ export declare function isOAuthToken(apiKey: string): boolean;
  * env-derived base; pass `undefined` to fall back to FOUNDRY/ANTHROPIC env
  * resolution and finally `DEFAULT_BASE_URL`.
  *
- * `isOAuth` is derived from the token prefix so the helper stays pure: callers
- * never have to thread the OAuth flag through their own resolution logic.
+ * `options.isOAuth` defaults to token detection. Pass the selected model's
+ * `isOAuth` to match the streaming path, which lets the model override the
+ * prefix (custom `anthropic-messages` providers default to OAuth shaping).
  */
-export declare function buildAnthropicAuthConfig(apiKey: string, baseUrl?: string): AnthropicAuthConfig;
+export declare function buildAnthropicAuthConfig(apiKey: string, baseUrl?: string, options?: AnthropicAuthOptions): AnthropicAuthConfig;
+/** Configured headers and fingerprint override policy for Anthropic hosted web search. */
+export interface AnthropicSearchHeaderOptions {
+    /** The selected model's configured headers (`ModelRegistry.resolveModelHeaders`). */
+    modelHeaders?: Record<string, string>;
+    /** The selected model's `compat.allowAnthropicHeaderOverrides`. */
+    allowAnthropicHeaderOverrides?: boolean;
+}
 /**
  * Builds HTTP headers for Anthropic API requests (search variant).
  *
- * Forwards `ANTHROPIC_CUSTOM_HEADERS` when the resolver deems them applicable
- * (Foundry mode, or a non-Anthropic base URL — typically an enterprise
- * gateway), matching the streaming path so web search behaves identically.
+ * Model headers and `ANTHROPIC_CUSTOM_HEADERS` (when the resolver deems them
+ * applicable: Foundry mode, or a non-Anthropic base URL such as an enterprise
+ * gateway) go through the same case-insensitive merge and enforced-header
+ * filtering as the streaming path, so web search behaves identically.
  */
-export declare function buildAnthropicSearchHeaders(auth: AnthropicAuthConfig): Record<string, string>;
+export declare function buildAnthropicSearchHeaders(auth: AnthropicAuthConfig, options?: AnthropicSearchHeaderOptions): Record<string, string>;
 /**
  * Builds the full API URL for Anthropic messages endpoint.
  */

package/dist/types/utils/resolve-peer.d.ts

@@ -0,0 +1,2 @@
+/** Use the socket peer unless the server explicitly trusts its reverse proxy. */
+export declare function resolvePeer(req: Request, socketAddress: string, trustProxyHeaders?: boolean): string;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-ai",
-	"version": "18.8.3",
+	"version": "18.8.5",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"keywords": [
 		"ai",
@@ -155,11 +155,11 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.8.3",
-		"@oh-my-pi/pi-catalog": "18.8.3",
-		"@oh-my-pi/pi-natives": "18.8.3",
-		"@oh-my-pi/pi-utils": "18.8.3",
-		"@oh-my-pi/pi-wire": "18.8.3"
+		"@oh-my-pi/omptype": "18.8.5",
+		"@oh-my-pi/pi-catalog": "18.8.5",
+		"@oh-my-pi/pi-natives": "18.8.5",
+		"@oh-my-pi/pi-utils": "18.8.5",
+		"@oh-my-pi/pi-wire": "18.8.5"
 	},
 	"devDependencies": {
 		"@types/bun": "^1.3.14"

package/src/auth/affinity.ts

@@ -360,19 +360,25 @@ export class SessionAffinity implements SessionsApi {
 	}
 
 	/**
-	 * Copy every stored credential affinity from one live session to another.
+	 * Copy stored credential affinities from one live session to another.
 	 *
 	 * The target receives its own sticky entries, so request resolution, usage
 	 * blocking, credential rotation, metadata, and persisted pins all continue
 	 * through the target session id without retaining a live dependency on the
-	 * source session.
+	 * source session. `include` limits the copy to the providers it accepts,
+	 * given whether the source's affinity is an explicit user pin.
 	 */
-	inherit(sourceSessionId: string, targetSessionId: string): number {
+	inherit(
+		sourceSessionId: string,
+		targetSessionId: string,
+		include?: (provider: string, explicit: boolean) => boolean,
+	): number {
 		if (!sourceSessionId || !targetSessionId || sourceSessionId === targetSessionId) return 0;
 		let inherited = 0;
 		for (const provider of this.#pool.providers()) {
 			const credential = this.get(provider, sourceSessionId);
 			if (!credential || !this.#permits(provider, targetSessionId, credential.index)) continue;
+			if (include && !include(provider, credential.explicit === true)) continue;
 			this.record(
 				provider,
 				targetSessionId,

package/src/auth/health.ts

@@ -1,5 +1,5 @@
 import type { Provider } from "../types";
-import { resolveUsedFraction } from "../usage";
+import { isWithinUsageReserve, resolveUsedFraction } from "../usage";
 import type {
 	CredentialRankingContext,
 	UsageCredential,
@@ -224,7 +224,7 @@ export class CredentialHealth implements HealthApi {
 				return {
 					credentialId: entry.id,
 					credentialType,
-					state: remainingFraction <= resolveReserveFraction(entry) ? "reserve" : "healthy",
+					state: isWithinUsageReserve(remainingFraction, resolveReserveFraction(entry)) ? "reserve" : "healthy",
 					remainingFraction,
 				};
 			}),

package/src/auth/oauth.ts

@@ -179,9 +179,20 @@ export class OAuthAccounts implements OAuthApi {
 		options: AuthApiKeyOptions | undefined,
 	): Promise<OAuthAccessResolution> {
 		try {
+			// tryOAuth refreshes only expired tokens; it resyncs this row from the store after a forced re-mint.
+			let index = selection.index;
+			if (options?.forceRefresh) {
+				await this.refresh(selection.credentialId, options.signal, {
+					reuseRecentMint: options.refreshReason === "auth-recovery",
+					reason: options.refreshReason,
+				});
+				// The refresh re-lists rows, so a concurrently removed row can shift this one's position.
+				index = this.#deps.pool.entries(provider).findIndex(entry => entry.id === selection.credentialId);
+				if (index === -1) throw new Error(`OAuth credential ${selection.credentialId} was removed during refresh`);
+			}
 			const resolved = await this.#deps.selector.tryOAuth(
 				provider,
-				{ credential: selection.credential, index: selection.index },
+				{ credential: selection.credential, index },
 				providerKey,
 				undefined,
 				options,
@@ -289,9 +300,10 @@ export class OAuthAccounts implements OAuthApi {
 	 * Resolve one stored OAuth credential by its durable storage row id.
 	 *
 	 * Unlike the normal session resolver, this method never ranks, rotates, or
-	 * falls back to sibling credentials. A forced refresh re-mints only the
-	 * requested row, preserving exact-account affinity for operations whose
-	 * provenance and policy boundary are tied to one workspace.
+	 * falls back to sibling credentials. A generic forced refresh re-mints only the
+	 * requested row; with `refreshReason: "auth-recovery"` it may instead reuse a
+	 * still-usable recent mint of that row. Either way exact-account affinity holds
+	 * for operations whose provenance and policy boundary are tied to one workspace.
 	 *
 	 * Returns `undefined` when the row does not exist for `provider` or an
 	 * explicit runtime/config API-key override suppresses OAuth.