Files
OmniRoute/open-sse/executors/base.ts
Diego Rodrigues de Sa e Souza f5073604b3 Release/v3.8.0 (#2430)
* fix(cli-tools): guard modelId type before calling indexOf

E2E shakedown v3.8.0: cli-tools quebrava com TypeError quando dynamicModels
continha entradas sem .id (objeto retornado diretamente em vez de string).

* fix(offline): avoid SSR/CSR hydration mismatch on navigator.onLine

Replace useState+lazy-initializer with useSyncExternalStore so the server
snapshot (() => false) and client snapshot (() => navigator.onLine) are
declared separately. React hydrates with the server value and switches to
the real online status client-side without a mismatch.

* chore(i18n): add missing en.json keys for translator, cli-tools, memory, onboarding

Adds 58 missing keys identified by the new dashboard audit script:
- cliTools: 18 custom CLI builder keys (CustomCliCard)
- translator: 24 keys covering stream transformer, live monitor, test bench
- memory: 12 health/pagination/dialog keys
- onboarding.tier: 8 keys for the tier tour walkthrough

Also adds scripts/i18n/audit-dashboard-pages.mjs which scans all dashboard
pages, reports t() calls referencing missing en.json keys, and flags
candidate hardcoded JSX/attribute strings.

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 1)

Subagents refactored 8 high-impact dashboard pages, replacing 81 of the
407 hardcoded English/PT strings flagged by the audit with proper
useTranslations() lookups. Added 73 corresponding keys to en.json across
the home, apiManager, providers, settings, and usage namespaces.

Pages affected:
- BudgetTab (27 → 0)
- HomePageClient (2 → 0)
- RoutingTab (25 → 7)
- ResilienceTab (38 → 18)
- SystemStorageTab (42 → 21)
- providers/[id] (17 → 15)
- ApiManagerPageClient (14 → 13)
- OneproxyTab (13 → 10)

Also adds two helper scripts:
- scripts/i18n/extract-keys-from-diff.mjs — extracts new keys from git diff
- scripts/i18n/merge-keys.mjs — merges a pending-keys JSON into en.json

Remaining hardcoded strings will be addressed in follow-up rounds.

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 2)

Continues round 1 (commit 8d34f4c65). Round-2 subagents refactored
additional dashboard pages, replacing 77 more hardcoded strings with
useTranslations() lookups. Added 79 corresponding keys to en.json
across the a2aDashboard, agents, analytics, apiManager, cliTools,
common, and settings namespaces.

Pages affected:
- a2a/page (new useTranslations + 6 keys)
- agent-skills/page (new useTranslations + 9 keys)
- AutoRoutingAnalyticsTab (new useTranslations + 6 keys)
- AppearanceTab (8 → 6 remaining)
- OneproxyTab (10 → 0)
- ResilienceTab (18 → 0 missing key)
- RoutingTab (7 → 0 missing key)
- VisionBridgeSettingsTab (new useTranslations + 6 keys)
- CopilotToolCard (7 → 0 missing key)
- ApiManagerPageClient (13 → 0 missing key)
- gamification/admin (new useTranslations + 7 keys)

Hardcoded total: 326 → 249. Real missing keys: 0 (the 6 still flagged
are false positives in exampleTemplates.tsx where t is passed as a
parameter — keys exist at translator.templatePayloads.*).

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 3)

Round-3 subagents and manual edits refactored 9 more dashboard pages
(plus 2 small extras), replacing ~80 hardcoded strings with
useTranslations() lookups. Added 79 corresponding keys to en.json
across analytics, cloudAgents, combos, common, health, settings, and
usage namespaces.

Pages affected:
- analytics/ComboHealthTab (new useTranslations + 15 keys)
- analytics/CompressionAnalyticsTab (new useTranslations + 11 keys)
- settings/SystemStorageTab (21 → 0 missing key)
- tokens/page (new useTranslations + 13 keys)
- usage/BudgetTab (9 missing fixed)
- health/page (manual: 6 keys)
- cloud-agents/page (manual: 3 keys)
- combos/page (manual: 1 key)

Hardcoded total: 249 → 164. Real missing keys: 0 (6 remaining are
exampleTemplates.tsx false positives).

Also adds scripts/i18n/build-pending-from-missing.mjs which reads
_audit.json and locates English values from HEAD to rebuild
_pending-keys.json after race-condition resets between subagent edits.

* chore(i18n): localize remaining dashboard settings labels

Replace hardcoded labels in compression and resilience settings with
translation lookups to continue the dashboard i18n cleanup.

Add the v3.8.0 dashboard shakedown runbook to document the manual
smoke-test process and known dev environment pitfalls.

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 4)

Round-4 subagent + manual key-resolution refactored remaining strings in
3 high-traffic settings/API tabs, plus extracted English values for
keys that were already added as t() calls but lost during the previous
en.json race-condition resets.

Pages affected:
- api-manager/ApiManagerPageClient (7 → 0 missing key)
- settings/CompressionSettingsTab (8 → 0 missing key)
- settings/MemorySkillsTab (8 → 0 missing key)
- settings/ResilienceTab (4 more keys recovered)

Hardcoded total: 164 → 140. Real missing keys: 0 (6 remaining are the
exampleTemplates.tsx false positives — t passed as parameter).

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 5)

Round-5 agent began processing the remaining smaller dashboard files.
Added 5 more keys to en.json for providers/[id]/page.tsx OAuth flow
labels and the cross-OS auto-detection hint.

Pages affected:
- providers/[id]/page.tsx (5 keys)

Hardcoded total: 140 → 136. Real missing keys: 0.

* chore(i18n): resolve last 2 missing providers/[id] keys

Adds providerDetailMyClaudeAccountPlaceholder and
providerDetailPathAutoDetected — the final user-visible labels in the
providers/[id] page that the round-5 subagent rewrote to t() calls
without yet adding to en.json.

Real missing keys: 0 (6 remaining are exampleTemplates.tsx false
positives — t is passed as a parameter so the audit cannot resolve the
namespace; keys do exist at translator.templatePayloads.*).

* chore(i18n): replace hardcoded UI text with t() calls across dashboard (round 6 — 10 parallel agents)

Round-6 dispatched 10 parallel subagents covering all 57 remaining
dashboard files. Each agent worked on a disjoint file set to avoid
en.json race conditions. Added ~60 new i18n keys across 9 namespaces
covering small UI labels, table headers, search placeholders, and
empty-state messages.

Major changes:
- analytics: SearchAnalyticsTab, ProviderUtilizationTab, DiversityScoreCard, CompressionAnalyticsTab (new useTranslations + keys)
- batch: BatchDetailModal, BatchListTab, FileDetailModal, FilesListTab (new useTranslations + keys)
- settings: CliproxyapiSettingsTab, PayloadRulesTab, ModelCooldownsCard, AppearanceTab, PricingTab (mostly new useTranslations)
- endpoint: TokenSaverCard, ApiEndpointsTab, EndpointPageClient
- cache: CachePerformance, IdempotencyLayer, ReasoningCacheTab, MediaPageClient, page
- combos: IntelligentComboPanel, page
- playground: ChatPlayground, SearchPlayground
- providers: ProviderCard
- onboarding: TierFlowDiagram
- changelog: ChangelogViewer
- home: ProviderTopology, TierCoverageWidget, BootstrapBanner, BadgeToast
- usage: BudgetTab, BudgetTelemetryCards, QuotaTable
- quotaShare: QuotaSharePageClient
- profile: page
- leaderboard: page
- skills: page

Hardcoded total: 131 → 60. Real missing keys: 0 plus 1 false-positive
for combos.modePack (lookup via prop-passed t).

* chore(i18n): finalize round-6 keys for batch/cache/endpoint/usage

Adds the remaining keys produced by parallel agents A4, A6, A8, A9:
- common: batch-related labels (BatchDetailModal, BatchListTab,
  FileDetailModal, FilesListTab, page) + profile/leaderboard
- cache: hit rate, latency, retry, avg chars
- endpoint: token saver, API endpoints, copy URL, cloud/local labels
- usage: noSpend, activeSessions, quotaAlerts, budget timing
- skills: install/marketplace/filter
- proxyRegistry/quotaShare/mcpDashboard: misc labels

Hardcoded total: 60 → 48. Real missing keys: 0 (modePack remaining is a
false positive — combos.modePack exists but the audit can't resolve it
since IntelligentComboPanel receives t as a prop).

* fix(playground): dedupe filteredModels to avoid duplicate React key warning

The /v1/models endpoint can return the same model id twice (e.g., when a
model is listed by both an alias and its canonical provider), which made
the <Select> emit two <option> elements with the same key — triggering
"Encountered two children with the same key, codex/gpt-5.5".

Replace the chained filter + map with a single pass that skips ids
already added.

* fix(playground): guard against non-string model ids before .split/.startsWith

The /v1/models endpoint can include synthetic entries (combos, locals,
in-progress imports) with a null/undefined id. The playground used to
call m.id.split("/") in the provider-discovery loop, which threw on the
first non-string entry; the surrounding .catch(() => {}) silently
swallowed the error, so the provider/model/account dropdowns ended up
empty even though /v1/models returned thousands of valid entries.

- Skip entries without a string id before split/startsWith.
- Log the rejection in the .catch handler so future regressions are
  visible in DevTools instead of silently emptying the UI.

* fix(playground): guard ChatPlayground filteredModels for non-string ids

Same root cause as commit 49fe356b9: ChatPlayground filtered models
with m.id.startsWith(...) which crashed on null/undefined ids returned
by /v1/models (synthetic combo entries). Apply the same defensive guard
and dedupe used in the parent page.

* fix(claude): drop orphan tool_result after fixToolAdjacency strip (discussion #2410)

Discussion #2410 reports Claude returning 400 for sequences like:
  assistant: tool_use(id=X)
  user: <plain text>           ← breaks adjacency
  user: tool_result(id=X)

The previous round added `fixToolAdjacency` (commit 44d9abac9) which
correctly strips the orphan tool_use from the assistant message. But
that left the now-unmatched tool_result intact, so the upstream
rejected the request with:

  messages.N.content.M: unexpected `tool_use_id` found in `tool_result`
  blocks: X. Each tool_result block must have a corresponding tool_use
  block in the previous message.

Fix: after running `fixToolAdjacency`, re-run `fixToolPairs` to drop
the orphaned tool_result blocks. All three call sites updated:
  - contextManager.purifyHistory (both inside the binary-search loop
    and the final pass)
  - BaseExecutor message-prep (Claude path)
  - claudeCodeCompatible request signer

Also tightens an unrelated dynamic-key access in
readNestedString (claudeCodeCompatible) to satisfy the prototype-
pollution scanner triggered by the post-tool semgrep hook.

* fix(mitm): point runtime manager re-export to js entrypoint

Use the emitted `.js` path for the runtime manager re-export so dynamic
runtime loading resolves correctly outside the Turbopack alias handling.

* docs: add AgentRouter setup guide (#2422)

Integrated into release/v3.8.0 — AgentRouter setup guide docs.

* feat: add new feature on combos - falloverBeforeRetry (#2417)

Integrated into release/v3.8.0 — falloverBeforeRetry for per-model quota skipping in combos.

* feat(batch): implement 10 feature requests harvested  (#2414)

Integrated into release/v3.8.0 — batch of 10 feature requests: llama.cpp local provider, upstream error exposure, Termux detection, providers rotate CLI, t3.chat web skeleton, Zed Docker integration, Kiro multi-account OAuth isolation, auto-combo cost blending, auto-combo context filter, combo provider-level exhaustion tracking (#1731). Conflicts with #2417 (falloverBeforeRetry) resolved.

* fix(gamification): resolve SQL bug, auth gap, pagination, and anomaly scoring (#2421)

Integrated into release/v3.8.0 — 6 critical gamification bug fixes: SQL SELECT in checkActionCountBadges, federation auth enforcement, leaderboard pagination offset, real z-score computation, addXp level calculation, and barrel index.ts

* docs(changelog): add post-release entries for #2414 #2417 #2421 #2422

- feat(batch): T3-Chat-Web executor, exhaustedProviders set (#1731), Zed Docker
- feat(combos): falloverBeforeRetry + setTry loop (#2417 — @hartmark)
- fix(gamification): SQL SELECT bug, federation auth, pagination, z-score (#2421 — @oyi77)
- docs: AgentRouter setup guide (#2422 — @leninejunior)

* fix(security): resolve CodeQL random/password-hash alerts and sync docs & tests

* feat/fix: integrate PRs #2423, #2425, #2427, #2428 with test & security fixes

* docs(changelog): credit contributors for PRs #2423, #2425, #2427, #2428

* fix(mitm): drop .js extension on manager.runtime re-export for webpack build (#2425)

Merged into release/v3.8.0

* fix: persist STORAGE_ENCRYPTION_KEY across upgrades (closes #1622) (#2428)

Merged into release/v3.8.0

* fix: auto-reset apiKeyHealth on successful connection test (#2427)

Merged into release/v3.8.0

* fix: support Antigravity image generation, Add Gemini 3.5 Flash (#2423)

Merged into release/v3.8.0

---------

Co-authored-by: diegosouzapw <diego.souza.pw@gmail.com>
Co-authored-by: Lenine Júnior <lenine@engrene.com.br>
Co-authored-by: Markus Hartung <mail@hartmark.se>
Co-authored-by: Paijo <14921983+oyi77@users.noreply.github.com>
Co-authored-by: Anton <39598727+NomenAK@users.noreply.github.com>
Co-authored-by: Chewji <126886556+Chewji9875@users.noreply.github.com>
Co-authored-by: clousky2020 <33016567+clousky2020@users.noreply.github.com>
Co-authored-by: backryun <bakryun0718@proton.me>
2026-05-20 09:12:49 -03:00

999 lines
38 KiB
TypeScript

import { HTTP_STATUS, FETCH_TIMEOUT_MS } from "../config/constants.ts";
import { applyFingerprint, isCliCompatEnabled } from "../config/cliFingerprints.ts";
import { supportsXHighEffort } from "../config/providerModels.ts";
import {
getRotatingApiKey,
getValidApiKey,
resolveKeyForRequest,
} from "../services/apiKeyRotator.ts";
import type { KeyHealth } from "../services/apiKeyRotator.ts";
import { getOpenAICompatibleType, isClaudeCodeCompatible } from "../services/provider.ts";
import type { ProviderRequestDefaults } from "../services/providerRequestDefaults.ts";
import { signRequestBody } from "../services/claudeCodeCCH.ts";
import {
appendAnthropicBetaHeader,
CONTEXT_1M_BETA_HEADER,
modelSupportsContext1mBeta,
} from "../services/claudeCodeCompatible.ts";
import { getClaudeCodeCompatibleRequestDefaults } from "@/lib/providers/requestDefaults";
import { remapToolNamesInRequest } from "../services/claudeCodeToolRemapper.ts";
import { obfuscateInBody } from "../services/claudeCodeObfuscation.ts";
import { applySystemTransformPipeline, PROVIDER_CLAUDE } from "../services/systemTransforms.ts";
import {
fixToolPairs,
fixToolAdjacency,
stripTrailingAssistantOrphanToolUse,
} from "../services/contextManager.ts";
import { randomUUID } from "node:crypto";
import {
CLAUDE_CODE_VERSION,
CLAUDE_CODE_STAINLESS_VERSION,
buildHashFor,
buildUserIdJson,
getSessionId,
parseUpstreamMetadataUserId,
passthroughUpstreamSessionId,
resolveAccountUUID,
resolveCliUserID,
selectBetaFlags,
stainlessArch,
stainlessOS,
stainlessRuntimeVersion,
stripProxyToolPrefix,
} from "./claudeIdentity.ts";
/**
* Sanitizes a custom API path to prevent path traversal attacks.
* Valid paths must start with '/', contain no '..' segments,
* no null bytes, and be reasonable in length.
*/
function sanitizePath(path: string): boolean {
if (typeof path !== "string") return false;
if (!path.startsWith("/")) return false;
if (path.includes("\0")) return false; // null byte
if (path.includes("..")) return false; // path traversal
if (path.length > 512) return false; // sanity limit
return true;
}
type JsonRecord = Record<string, unknown>;
export type ProviderConfig = {
id?: string;
baseUrl?: string;
baseUrls?: string[];
responsesBaseUrl?: string;
chatPath?: string;
clientVersion?: string;
clientId?: string;
clientSecret?: string;
tokenUrl?: string;
refreshUrl?: string;
authUrl?: string;
headers?: Record<string, string>;
requestDefaults?: ProviderRequestDefaults;
timeoutMs?: number;
format?: string;
};
export type ProviderCredentials = {
accessToken?: string;
refreshToken?: string;
apiKey?: string;
projectId?: string | null;
expiresAt?: string;
connectionId?: string; // T07: used for API key rotation index
maxConcurrent?: number | null;
providerSpecificData?: JsonRecord;
requestEndpointPath?: string;
};
export type ExecutorLog = {
debug?: (tag: string, message: string) => void;
info?: (tag: string, message: string) => void;
warn?: (tag: string, message: string) => void;
error?: (tag: string, message: string) => void;
};
export type ExecuteInput = {
model: string;
body: unknown;
stream: boolean;
credentials: ProviderCredentials;
signal?: AbortSignal | null;
log?: ExecutorLog | null;
extendedContext?: boolean;
/** Merged after auth + CLI fingerprint headers (values override same-named defaults). */
upstreamExtraHeaders?: Record<string, string> | null;
/** Original client request headers (read-only). Executors may forward select headers upstream. */
clientHeaders?: Record<string, string> | null;
/** Callback to persist tokens that are proactively refreshed during execution. */
onCredentialsRefreshed?: (newCredentials: ProviderCredentials) => Promise<void> | void;
/** When true, skip the intra-URL 429 retry in execute() so the caller handles fallback. */
skipUpstreamRetry?: boolean;
};
export type CountTokensInput = {
body: Record<string, unknown>;
credentials: ProviderCredentials;
log?: ExecutorLog | null;
model: string;
signal?: AbortSignal | null;
};
/** Apply model-level extra upstream headers (e.g. Authentication, X-Custom-Auth). */
export function mergeUpstreamExtraHeaders(
headers: Record<string, string>,
extra?: Record<string, string> | null
): void {
if (!extra) return;
for (const [k, v] of Object.entries(extra)) {
if (typeof k === "string" && k.length > 0 && typeof v === "string") {
if (k.toLowerCase() === "user-agent") {
setUserAgentHeader(headers, v);
continue;
}
headers[k] = v;
}
}
}
export function getCustomUserAgent(providerSpecificData?: JsonRecord | null): string | null {
const customUserAgent =
typeof providerSpecificData?.customUserAgent === "string"
? providerSpecificData.customUserAgent.trim()
: "";
return customUserAgent || null;
}
export function setUserAgentHeader(headers: Record<string, string>, userAgent: string): void {
headers["User-Agent"] = userAgent;
if ("user-agent" in headers) {
headers["user-agent"] = userAgent;
}
}
export function applyConfiguredUserAgent(
headers: Record<string, string>,
providerSpecificData?: JsonRecord | null
): void {
const customUserAgent = getCustomUserAgent(providerSpecificData);
if (customUserAgent) {
setUserAgentHeader(headers, customUserAgent);
}
}
export function mergeAbortSignals(primary: AbortSignal, secondary: AbortSignal): AbortSignal {
const controller = new AbortController();
const abortFrom = (source: AbortSignal) => {
if (!controller.signal.aborted) {
controller.abort(source.reason);
}
};
if (primary.aborted) {
abortFrom(primary);
return controller.signal;
}
if (secondary.aborted) {
abortFrom(secondary);
return controller.signal;
}
primary.addEventListener("abort", () => abortFrom(primary), { once: true });
secondary.addEventListener("abort", () => abortFrom(secondary), { once: true });
return controller.signal;
}
/**
* Sanitize reasoning_effort for providers that don't accept all values.
*
* The claude→openai translator emits reasoning_effort=xhigh when the client
* sends output_config.effort=max on a Claude-shape request. Combined with
* runtime alias remapping (e.g. claude-opus-4-6 → mimo/mimo-v2.5-pro), this
* routes xhigh to OpenAI-shape providers that don't accept the value:
*
* xiaomi-mimo : low|medium|high only — 400 literal_error on xhigh
* mistral : devstral models reject reasoning_effort entirely
* github : claude/haiku/oswe models reject reasoning_effort entirely
*
* Each rejection burns a combo fallback attempt before reaching a working
* provider. Apply provider-aware sanitation here (after transformRequest, so
* reintroductions by per-provider transforms are also caught) before fetch.
* Models that genuinely support xhigh (registry flag supportsXHighEffort)
* pass through unchanged.
*/
const MISTRAL_NO_REASONING_EFFORT_PATTERN = /devstral/i;
const GITHUB_NO_REASONING_EFFORT_PATTERN = /(claude|haiku|oswe)/i;
export function sanitizeReasoningEffortForProvider(
body: unknown,
provider: string,
model: string | undefined,
log?: { info?: (tag: string, msg: string) => void } | null
): unknown {
if (!body || typeof body !== "object" || Array.isArray(body)) return body;
const b = body as Record<string, unknown>;
const reasoning =
b.reasoning && typeof b.reasoning === "object" && !Array.isArray(b.reasoning)
? (b.reasoning as Record<string, unknown>)
: null;
const effort = b.reasoning_effort ?? reasoning?.effort;
if (effort === undefined) return body;
const effortStr = typeof effort === "string" ? effort.toLowerCase() : "";
const modelStr = model || "";
if (effortStr === "xhigh" && !supportsXHighEffort(provider, modelStr)) {
log?.info?.(
"REASONING_SANITIZE",
`${provider}/${modelStr}: downgraded reasoning_effort xhigh → high`
);
const next: Record<string, unknown> = { ...b, reasoning_effort: "high" };
if (reasoning) {
next.reasoning = { ...reasoning, effort: "high" };
}
return next;
}
const rejecting =
(provider === "mistral" && MISTRAL_NO_REASONING_EFFORT_PATTERN.test(modelStr)) ||
(provider === "github" && GITHUB_NO_REASONING_EFFORT_PATTERN.test(modelStr));
if (rejecting) {
log?.info?.(
"REASONING_SANITIZE",
`${provider}/${modelStr}: removed unsupported reasoning_effort`
);
const next: Record<string, unknown> = { ...b };
delete next.reasoning_effort;
if (reasoning) {
const r = { ...reasoning };
delete r.effort;
if (Object.keys(r).length === 0) delete next.reasoning;
else next.reasoning = r;
}
return next;
}
return body;
}
/**
* BaseExecutor - Base class for provider executors.
* Implements the Strategy pattern: subclasses override specific methods
* (buildUrl, buildHeaders, transformRequest, etc.) for each provider.
*/
export class BaseExecutor {
provider: string;
config: ProviderConfig;
constructor(provider: string, config: ProviderConfig) {
this.provider = provider;
this.config = config;
}
getProvider() {
return this.provider;
}
getBaseUrls() {
return this.config.baseUrls || (this.config.baseUrl ? [this.config.baseUrl] : []);
}
getFallbackCount() {
return this.getBaseUrls().length || 1;
}
getTimeoutMs() {
const configured = this.config?.timeoutMs;
if (typeof configured !== "number" || !Number.isFinite(configured)) {
return FETCH_TIMEOUT_MS;
}
return Math.max(1, Math.floor(configured));
}
getCountTokensTimeoutMs() {
return this.getTimeoutMs();
}
buildUrl(
model: string,
stream: boolean,
urlIndex = 0,
credentials: ProviderCredentials | null = null
) {
void model;
void stream;
if (this.provider?.startsWith?.("openai-compatible-")) {
const psd = credentials?.providerSpecificData;
const baseUrl = typeof psd?.baseUrl === "string" ? psd.baseUrl : "https://api.openai.com/v1";
const normalized = baseUrl.replace(/\/$/, "");
// Sanitize custom path: must start with '/', no path traversal, no null bytes
const rawPath = typeof psd?.chatPath === "string" && psd.chatPath ? psd.chatPath : null;
const customPath = rawPath && sanitizePath(rawPath) ? rawPath : null;
if (customPath) return `${normalized}${customPath}`;
const path =
getOpenAICompatibleType(this.provider, psd) === "responses"
? "/responses"
: "/chat/completions";
return `${normalized}${path}`;
}
const baseUrls = this.getBaseUrls();
return baseUrls[urlIndex] || baseUrls[0] || this.config.baseUrl || "";
}
buildHeaders(
credentials: ProviderCredentials,
stream = true,
clientHeaders?: Record<string, string> | null,
model?: string,
health?: Record<string, KeyHealth>
): Record<string, string> {
void clientHeaders;
void model;
const headers: Record<string, string> = {
"Content-Type": "application/json",
...this.config.headers,
};
// Allow per-provider User-Agent override via environment variable.
// Example: CLAUDE_USER_AGENT="my-agent/2.0" overrides the default for the Claude provider.
const providerId = this.config?.id || this.provider;
if (providerId) {
const envKey = `${providerId.toUpperCase().replace(/[^A-Z0-9]/g, "_")}_USER_AGENT`;
const envUA = process.env[envKey]?.trim();
if (envUA) {
setUserAgentHeader(headers, envUA);
}
}
if (credentials.accessToken) {
headers["Authorization"] = `Bearer ${credentials.accessToken}`;
} else if (credentials.apiKey) {
const extraKeys =
(credentials.providerSpecificData?.extraApiKeys as string[] | undefined) ?? [];
const selectedKeyId = (
credentials.providerSpecificData as Record<string, unknown> | undefined
)?.selectedKeyId as string | undefined;
let effectiveKey = credentials.apiKey;
if (extraKeys.length > 0 && credentials.connectionId) {
const resolved = resolveKeyForRequest(
credentials.connectionId,
credentials.apiKey,
extraKeys,
selectedKeyId ?? null
);
effectiveKey = resolved?.key ?? credentials.apiKey;
if (resolved && credentials.providerSpecificData) {
(credentials.providerSpecificData as Record<string, unknown>).selectedKeyId =
resolved.keyId;
}
}
headers["Authorization"] = `Bearer ${effectiveKey}`;
}
headers["Accept"] = stream ? "text/event-stream" : "application/json";
return headers;
}
// Override in subclass for provider-specific transformations
transformRequest(
model: string,
body: unknown,
stream: boolean,
credentials: ProviderCredentials
): unknown {
void model;
void stream;
void credentials;
// Fix #1674: Remove empty string values from optional parameters
// like tool descriptions to avoid upstream validation failures.
if (body && typeof body === "object" && !Array.isArray(body)) {
const cloned = { ...body } as Record<string, unknown>;
if (Array.isArray(cloned.tools)) {
cloned.tools = cloned.tools.map((tool: unknown) => {
if (tool && typeof tool === "object" && !Array.isArray(tool)) {
const toolRecord = tool as JsonRecord;
const toolFunction = toolRecord.function;
if (toolFunction && typeof toolFunction === "object" && !Array.isArray(toolFunction)) {
const func = { ...(toolFunction as JsonRecord) };
if (func.description === "") delete func.description;
if (typeof func.name !== "string" || func.name.trim() === "") {
func.name = "unnamed_tool";
}
return { ...toolRecord, function: func };
}
}
return tool;
});
}
// Fix #1884: Cursor sends prompt_cache_retention which breaks strict upstream endpoints
delete cloned.prompt_cache_retention;
// Also clean up top level optional fields that commonly cause issues when empty
const optionalKeys = ["user", "stop", "seed", "response_format"];
for (const key of optionalKeys) {
if (cloned[key] === "") delete cloned[key];
}
return cloned;
}
return body;
}
shouldRetry(status: number, urlIndex: number) {
return status === HTTP_STATUS.RATE_LIMITED && urlIndex + 1 < this.getFallbackCount();
}
// Intra-URL retry config: retry same URL before falling back to next node
static readonly RETRY_CONFIG = { maxAttempts: 2, delayMs: 2000 };
// Timeout for receiving the initial upstream response headers. Once the response
// starts streaming, STREAM_IDLE_TIMEOUT_MS / Undici bodyTimeout handle stalls.
static FETCH_START_TIMEOUT_MS = FETCH_TIMEOUT_MS;
// Override in subclass for provider-specific refresh
async refreshCredentials(
credentials: ProviderCredentials,
log: ExecutorLog | null
): Promise<Partial<ProviderCredentials> | null> {
void credentials;
void log;
return null;
}
needsRefresh(credentials?: ProviderCredentials | null) {
if (!credentials?.expiresAt) return false;
const expiresAtMs = new Date(credentials.expiresAt).getTime();
return expiresAtMs - Date.now() < 5 * 60 * 1000;
}
parseError(response: Response, bodyText: string) {
return { status: response.status, message: bodyText || `HTTP ${response.status}` };
}
buildCountTokensUrl(model: string, credentials: ProviderCredentials | null = null) {
void model;
void credentials;
const baseUrl = this.buildUrl(model, false, 0, credentials);
if (typeof baseUrl !== "string" || baseUrl.length === 0) return null;
if (this.config?.format !== "claude" || !baseUrl.includes("/messages")) return null;
const [path, query = ""] = baseUrl.split("?");
const normalizedPath = path.endsWith("/messages")
? `${path}/count_tokens`
: `${path}/count_tokens`;
return query ? `${normalizedPath}?${query}` : normalizedPath;
}
async countTokens({ model, body, credentials, signal, log }: CountTokensInput) {
const url = this.buildCountTokensUrl(model, credentials);
if (!url) return null;
const headers = this.buildHeaders(credentials, false);
const requestBody =
body && typeof body === "object"
? {
...body,
model,
}
: { model };
let timeoutId: ReturnType<typeof setTimeout> | null = null;
let activeSignal = signal || null;
let controller: AbortController | null = null;
const timeoutMs = this.getCountTokensTimeoutMs();
if (timeoutMs > 0) {
controller = new AbortController();
timeoutId = setTimeout(() => controller?.abort(), timeoutMs);
activeSignal = signal ? mergeAbortSignals(signal, controller.signal) : controller.signal;
}
try {
const response = await fetch(url, {
method: "POST",
headers,
body: JSON.stringify(requestBody),
signal: activeSignal || undefined,
});
const text = await response.text();
if (!response.ok) {
const parsedError = this.parseError(response, text);
throw new Error(parsedError.message);
}
const parsed = text ? JSON.parse(text) : {};
const inputTokens = Number(parsed?.input_tokens);
if (!Number.isFinite(inputTokens)) {
throw new Error("Provider count_tokens response missing input_tokens");
}
return { input_tokens: inputTokens, provider: this.provider, source: "provider" };
} catch (error) {
log?.debug?.(
"COUNT_TOKENS",
`${this.provider}/${model} real count unavailable: ${error instanceof Error ? error.message : String(error)}`
);
return null;
} finally {
if (timeoutId) clearTimeout(timeoutId);
}
}
async execute({
model,
body,
stream,
credentials,
signal,
log,
extendedContext,
upstreamExtraHeaders,
clientHeaders,
skipUpstreamRetry = false,
}: ExecuteInput) {
const fallbackCount = this.getFallbackCount();
let lastError: unknown = null;
let lastStatus = 0;
let activeCredentials = credentials;
// Track per-URL intra-retry attempts to avoid infinite loops
const retryAttemptsByUrl: Record<number, number> = {};
if (this.needsRefresh(credentials)) {
try {
const refreshed = await this.refreshCredentials(credentials, log || null);
if (refreshed) {
activeCredentials = {
...credentials,
...refreshed,
};
// Persist the proactively refreshed credentials to prevent consuming rotating tokens
// without updating the central database connection.
if (arguments[0].onCredentialsRefreshed) {
await arguments[0].onCredentialsRefreshed(refreshed);
}
}
} catch (error) {
log?.warn?.(
"TOKEN",
`Credential refresh failed for ${this.provider}: ${error instanceof Error ? error.message : String(error)}`
);
}
}
for (let urlIndex = 0; urlIndex < fallbackCount; urlIndex++) {
const url = this.buildUrl(model, stream, urlIndex, activeCredentials);
const headers = this.buildHeaders(activeCredentials, stream, clientHeaders, model);
applyConfiguredUserAgent(headers, activeCredentials?.providerSpecificData);
const ccRequestDefaults = isClaudeCodeCompatible(this.provider)
? getClaudeCodeCompatibleRequestDefaults(activeCredentials?.providerSpecificData)
: {};
const shouldForwardExtendedContext =
extendedContext &&
modelSupportsContext1mBeta(model) &&
!isClaudeCodeCompatible(this.provider);
const shouldForwardCcCompatibleContext1m =
isClaudeCodeCompatible(this.provider) && ccRequestDefaults.context1m === true;
if (shouldForwardExtendedContext || shouldForwardCcCompatibleContext1m) {
appendAnthropicBetaHeader(headers, CONTEXT_1M_BETA_HEADER);
}
const rawTransformedBody = await this.transformRequest(
model,
body,
stream,
activeCredentials
);
const transformedBody = sanitizeReasoningEffortForProvider(
rawTransformedBody,
this.provider,
model,
log
);
try {
// Only enforce the timeout while waiting for the initial fetch() response.
// Once headers arrive, active streams must not be cut off by total elapsed time;
// post-start stalls are handled separately by STREAM_IDLE_TIMEOUT_MS / bodyTimeout.
const fetchStartTimeoutMs = this.getTimeoutMs();
const timeoutController = fetchStartTimeoutMs > 0 ? new AbortController() : null;
let timeoutId: ReturnType<typeof setTimeout> | null = null;
if (timeoutController) {
timeoutId = setTimeout(() => {
const timeoutError = new Error(
`Fetch timeout after ${fetchStartTimeoutMs}ms on ${url}`
);
timeoutError.name = "TimeoutError";
timeoutController.abort(timeoutError);
}, fetchStartTimeoutMs);
}
const timeoutSignal = timeoutController?.signal ?? null;
const combinedSignal =
signal && timeoutSignal
? mergeAbortSignals(signal, timeoutSignal)
: signal || timeoutSignal;
const isClaudeCodeClient =
clientHeaders?.["x-app"] === "cli" ||
(clientHeaders?.["user-agent"] &&
clientHeaders["user-agent"].toLowerCase().includes("claude-code")) ||
(clientHeaders?.["user-agent"] &&
clientHeaders["user-agent"].toLowerCase().includes("claude-cli"));
// Anthropic's user:sessions:claude_code OAuth scope expects CLI-shaped
// traffic. Apply the cloak whenever we have an OAuth token, regardless
// of upstream client.
const hasClaudeOAuthToken =
typeof activeCredentials?.accessToken === "string" &&
activeCredentials.accessToken.startsWith("sk-ant-oat") &&
!activeCredentials?.apiKey;
if (
this.provider === "claude" &&
(isClaudeCodeClient || hasClaudeOAuthToken) &&
typeof transformedBody === "object" &&
transformedBody !== null
) {
const tb = transformedBody as Record<string, unknown>;
stripProxyToolPrefix(tb);
remapToolNamesInRequest(tb);
obfuscateInBody(tb);
// NOTE (issue #2260): This is the native `claude` provider OAuth path.
// It is intentionally NOT routed through applyCcBridgeTransformPipeline.
// The native OAuth path already prepends its own billing line + sentinel
// (see lines ~744-773 below, dayStamp-based, cc_entrypoint=cli, cch=00000
// placeholder, signed at body level). The CC bridge transforms DSL is
// wired into buildAndSignClaudeCodeRequest (claudeCodeCompatible.ts step 5b)
// which is the anthropic-compatible-cc-* relay path — a different,
// separately classified surface. Do not double-prepend here.
// Real CLI never sets cache_control on tools.
if (Array.isArray(tb.tools)) {
for (const t of tb.tools as Array<Record<string, unknown>>) {
delete t.cache_control;
}
}
// Per-request behavior overrides via custom client headers.
// x-omniroute-effort: low | medium | high | xhigh | off
// x-omniroute-thinking: adaptive | off
// A header value applies only when the corresponding body field is
// not already set; "off" force-strips the field.
const headerEffort = (
clientHeaders?.["x-omniroute-effort"] ?? clientHeaders?.["X-OmniRoute-Effort"]
)
?.trim()
.toLowerCase();
const headerThinking = (
clientHeaders?.["x-omniroute-thinking"] ?? clientHeaders?.["X-OmniRoute-Thinking"]
)
?.trim()
.toLowerCase();
let appliedEffort: string | null = null;
let appliedThinking: string | null = null;
if (headerEffort === "off") {
if (tb.output_config && typeof tb.output_config === "object") {
delete (tb.output_config as Record<string, unknown>).effort;
}
appliedEffort = "off";
} else if (headerEffort && ["low", "medium", "high", "xhigh"].includes(headerEffort)) {
const oc =
tb.output_config && typeof tb.output_config === "object"
? (tb.output_config as Record<string, unknown>)
: {};
if (oc.effort === undefined) {
oc.effort = headerEffort;
tb.output_config = oc;
appliedEffort = headerEffort;
}
}
if (headerThinking === "adaptive") {
if (tb.thinking === undefined) {
tb.thinking = { type: "adaptive" };
appliedThinking = "adaptive";
}
if (tb.context_management === undefined) {
tb.context_management = {
edits: [{ type: "clear_thinking_20251015", keep: "all" }],
};
}
} else if (headerThinking === "off") {
delete tb.thinking;
delete tb.context_management;
appliedThinking = "off";
} else if (!headerThinking && !headerEffort) {
// Default CC logic when no override headers are present
const isHaiku = typeof tb.model === "string" && tb.model.includes("haiku");
if (isHaiku) {
delete tb.thinking;
delete tb.output_config;
delete tb.context_management;
} else if (tb.thinking === undefined && tb.output_config === undefined) {
tb.thinking = { type: "adaptive" };
tb.context_management = {
edits: [{ type: "clear_thinking_20251015", keep: "all" }],
};
tb.output_config = { effort: "high" };
}
}
// Real CLI always pairs context_management with thinking. Mirror
// that invariant so long sessions don't accumulate thinking blocks
// toward the context cap.
if (tb.thinking && !tb.context_management) {
tb.context_management = {
edits: [{ type: "clear_thinking_20251015", keep: "all" }],
};
}
const seed = activeCredentials?.accessToken || activeCredentials?.apiKey || "anon";
const psd = activeCredentials?.providerSpecificData as
| Record<string, unknown>
| undefined;
let identitySource:
| "upstream-metadata"
| "upstream-header"
| "synthesized"
| "synthesized-cloaked" = "synthesized";
let sessionId: string;
let deviceId: string;
let accountUUID: string;
// For any Claude OAuth request, ignore client-supplied metadata.user_id /
// X-Claude-Code-Session-Id and synthesize per-account: the CC device_id from
// ~/.claude.json is shared across every account on one machine, which lets
// Anthropic correlate accounts behind one OmniRoute.
const cloakIdentity = isClaudeCodeClient || hasClaudeOAuthToken;
const upstreamUserId = cloakIdentity ? null : parseUpstreamMetadataUserId(tb);
if (upstreamUserId) {
sessionId = upstreamUserId.session_id;
deviceId = upstreamUserId.device_id;
accountUUID = upstreamUserId.account_uuid;
identitySource = "upstream-metadata";
} else {
const headerSid = cloakIdentity
? null
: passthroughUpstreamSessionId(
clientHeaders as Record<string, string | undefined> | undefined
);
sessionId = headerSid ?? getSessionId(seed);
deviceId = resolveCliUserID(psd, seed);
accountUUID = resolveAccountUUID(psd, seed, activeCredentials?.accessToken);
identitySource = headerSid
? "upstream-header"
: cloakIdentity
? "synthesized-cloaked"
: "synthesized";
}
// system[0] (billing) and system[1] (sentinel) must not carry
// cache_control — that belongs on upstream prompt blocks at [2..].
const dayStamp = new Date().toISOString().slice(0, 10);
const buildHash = buildHashFor(CLAUDE_CODE_VERSION, dayStamp);
const billingLine = `x-anthropic-billing-header: cc_version=${CLAUDE_CODE_VERSION}.${buildHash}; cc_entrypoint=cli; cch=00000;`;
const SENTINEL = "You are Claude Code, Anthropic's official CLI for Claude.";
const sysBlocks: Array<Record<string, unknown>> = Array.isArray(tb.system)
? (tb.system as Array<Record<string, unknown>>)
: typeof tb.system === "string"
? [{ type: "text", text: tb.system }]
: [];
// Strip any pre-existing billing/sentinel before re-prepending — keeps
// retries idempotent and avoids stacking that breaks prompt-cache prefix
// matching (see issue #1712).
for (let i = sysBlocks.length - 1; i >= 0; i--) {
const t = sysBlocks[i]?.text;
if (typeof t === "string" && t.startsWith("x-anthropic-billing-header:")) {
sysBlocks.splice(i, 1);
}
}
for (let i = sysBlocks.length - 1; i >= 0; i--) {
const t = sysBlocks[i]?.text;
if (typeof t === "string" && t.startsWith(SENTINEL)) {
sysBlocks.splice(i, 1);
}
}
sysBlocks.unshift({ type: "text", text: billingLine }, { type: "text", text: SENTINEL });
tb.system = sysBlocks;
// Run the configurable system-transforms pipeline for the native
// `claude` provider (issue #2260 / comment 4459544580). The default
// claude pipeline runs cosmetic ops only (Open WebUI paragraph
// anchors, identity-prefix paragraph drop, ZWJ obfuscation of
// sensitive words). It deliberately does NOT include
// `inject_billing_header` — billing + sentinel are already
// prepended above. Users can extend the pipeline via Settings UI.
{
const transformResult = applySystemTransformPipeline(PROVIDER_CLAUDE, tb);
if (transformResult.appliedOpKinds.length > 0) {
console.log(
`[SystemTransforms] claude-native: ${transformResult.appliedOpKinds.join(", ")}`
);
}
}
if (!tb.metadata || typeof tb.metadata !== "object") tb.metadata = {};
(tb.metadata as Record<string, unknown>).user_id = buildUserIdJson({
deviceId,
accountUUID,
sessionId,
});
// Headers. Accept stays application/json even on streams (Stainless
// convention; SSE decoding is gated on body.stream). anthropic-beta
// is selected per request shape; the full set on a quota probe is
// itself a fingerprint.
const ccHeaders: Record<string, string> = {
Accept: "application/json",
"anthropic-version": "2023-06-01",
"anthropic-beta": selectBetaFlags(tb),
"anthropic-dangerous-direct-browser-access": "true",
"x-app": "cli",
"User-Agent": `claude-cli/${CLAUDE_CODE_VERSION} (external, cli)`,
"X-Stainless-Package-Version": CLAUDE_CODE_STAINLESS_VERSION,
"X-Stainless-Timeout": "600",
"accept-encoding": "gzip, deflate, br, zstd",
connection: "keep-alive",
"x-client-request-id": randomUUID(),
"X-Claude-Code-Session-Id": sessionId,
};
// Drop case variants of the same header name before merging — undici
// would otherwise concatenate them (issue #1454).
const ccKeysLower = new Set(Object.keys(ccHeaders).map((k) => k.toLowerCase()));
for (const key of Object.keys(headers)) {
if (ccKeysLower.has(key.toLowerCase())) delete headers[key];
}
Object.assign(headers, ccHeaders);
delete headers["X-Stainless-Helper-Method"];
// Stainless OS/Arch/Runtime are host-derived (Stainless SDK does the
// same at runtime). Hardcoding them was a unique-per-deployment tell.
headers["X-Stainless-Arch"] = stainlessArch();
headers["X-Stainless-Lang"] = "js";
headers["X-Stainless-OS"] = stainlessOS();
headers["X-Stainless-Runtime"] = "node";
headers["X-Stainless-Runtime-Version"] = stainlessRuntimeVersion();
headers["X-Stainless-Retry-Count"] = "0";
delete headers["X-Stainless-Os"];
const overrideTag =
appliedEffort || appliedThinking
? ` overrides=effort:${appliedEffort ?? "-"},thinking:${appliedThinking ?? "-"}`
: "";
log?.debug?.(
"CLAUDE",
`identity=${identitySource} sid=${sessionId.slice(0, 8)} dev=${deviceId.slice(0, 8)} acct=${accountUUID.slice(0, 8)}${overrideTag}`
);
}
// CLI fingerprint ordering — always-on for native Claude OAuth, opt-in
// for other providers. Header + body field order is itself a fingerprint.
let finalHeaders = headers;
// Strip internal sentinel fields set by remapToolNamesInRequest before
// serializing — Anthropic rejects unknown top-level fields (issue #2260).
delete (transformedBody as Record<string, unknown>)[
"_claudeCodeRequiresLowercaseToolNames"
];
// Guard against orphan tool_use / tool_result pairs. Clients can ship
// truncated histories mid-tool-call which Anthropic rejects with
// `messages.N: tool_use ids were found without tool_result blocks
// immediately after: toolu_...`. fixToolPairs strips orphans, then
// stripTrailingAssistantOrphanToolUse catches the case where the
// request body itself ends on an unmatched assistant(tool_use) —
// invalid for an upstream-send turn since the body must end on a
// user message. Both are idempotent on clean histories.
{
const tb = transformedBody as Record<string, unknown>;
if (Array.isArray(tb?.messages)) {
const fixed = fixToolPairs(tb.messages as Record<string, unknown>[]);
// fixToolAdjacency enforces Claude's strict adjacency rule
// (tool_result must be in immediately next message).
// Only apply for Claude/Claude-compatible — OpenAI allows results
// spread across multiple subsequent messages.
const isClaude = this.provider === "claude" || isClaudeCodeCompatible(this.provider);
// For Claude, fixToolAdjacency may strip tool_use blocks whose
// tool_result isn't in the next message; re-run fixToolPairs to
// drop any tool_result orphaned by that strip (discussion #2410).
const adjacent = isClaude ? fixToolPairs(fixToolAdjacency(fixed)) : fixed;
tb.messages = stripTrailingAssistantOrphanToolUse(adjacent);
}
}
let bodyString = JSON.stringify(transformedBody);
const shouldFingerprint =
isCliCompatEnabled(this.provider) ||
(this.provider === "claude" && (isClaudeCodeClient || hasClaudeOAuthToken));
if (shouldFingerprint) {
const fingerprinted = applyFingerprint(this.provider, headers, transformedBody);
finalHeaders = fingerprinted.headers;
bodyString = fingerprinted.bodyString;
}
// CCH signing — replaces the cch=00000 placeholder in the billing
// header with an xxHash64 integrity token over the serialized body.
if (isClaudeCodeCompatible(this.provider) || this.provider === "claude") {
bodyString = await signRequestBody(bodyString);
}
mergeUpstreamExtraHeaders(finalHeaders, upstreamExtraHeaders);
const fetchOptions: RequestInit = {
method: "POST",
headers: finalHeaders,
body: bodyString,
};
if (combinedSignal) fetchOptions.signal = combinedSignal;
let response;
try {
response = await fetch(url, fetchOptions);
} finally {
if (timeoutId) {
clearTimeout(timeoutId);
timeoutId = null;
}
}
// Intra-URL retry: if 429 and we haven't exhausted per-URL retries, wait and retry the same URL
if (
!skipUpstreamRetry &&
response.status === HTTP_STATUS.RATE_LIMITED &&
(retryAttemptsByUrl[urlIndex] ?? 0) < BaseExecutor.RETRY_CONFIG.maxAttempts
) {
retryAttemptsByUrl[urlIndex] = (retryAttemptsByUrl[urlIndex] ?? 0) + 1;
const attempt = retryAttemptsByUrl[urlIndex];
log?.debug?.(
"RETRY",
`429 intra-retry ${attempt}/${BaseExecutor.RETRY_CONFIG.maxAttempts} on ${url} — waiting ${BaseExecutor.RETRY_CONFIG.delayMs}ms`
);
await new Promise((resolve) => setTimeout(resolve, BaseExecutor.RETRY_CONFIG.delayMs));
urlIndex--; // re-run this urlIndex on the next loop iteration
continue;
}
// T07: Handle 401 authentication errors — log and continue to fallback
if (response.status === 401 && credentials.connectionId && credentials.apiKey) {
log?.warn?.("AUTH", `401 on ${url} - API key may be invalid`);
}
if (!skipUpstreamRetry && this.shouldRetry(response.status, urlIndex)) {
log?.debug?.("RETRY", `${response.status} on ${url}, trying fallback ${urlIndex + 1}`);
lastStatus = response.status;
continue;
}
return { response, url, headers: finalHeaders, transformedBody };
} catch (error) {
// Distinguish timeout errors from other abort errors
const err = error instanceof Error ? error : new Error(String(error));
if (err.name === "TimeoutError") {
log?.warn?.("TIMEOUT", `Fetch timeout after ${this.getTimeoutMs()}ms on ${url}`);
}
lastError = err;
if (!skipUpstreamRetry && urlIndex + 1 < fallbackCount) {
log?.debug?.("RETRY", `Error on ${url}, trying fallback ${urlIndex + 1}`);
continue;
}
throw err;
}
}
throw lastError || new Error(`All ${fallbackCount} URLs failed with status ${lastStatus}`);
}
}
export default BaseExecutor;