OmniRoute/open-sse/executors/codex.ts

import { getCodexRequestDefaults } from "@/lib/providers/requestDefaults";
import { BaseExecutor, setUserAgentHeader } from "./base.ts";
import { CODEX_DEFAULT_INSTRUCTIONS } from "../config/codexInstructions.ts";
import { PROVIDERS } from "../config/constants.ts";
import { getCodexClientVersion, getCodexUserAgent } from "../config/codexClient.ts";
import { getAccessToken } from "../services/tokenRefresh.ts";
import { getThinkingBudgetConfig, ThinkingMode } from "../services/thinkingBudget.ts";

// ─── T09: Codex vs Spark Scope-Aware Rate Limiting ────────────────────────
// Codex has two independent quota pools: "codex" (standard) and "spark" (premium).
// Exhausting one should NOT block requests to the other.
// Ref: sub2api PR #1129 (feat(openai): split codex spark rate limiting from codex)

/**
 * Maps model name substrings to their rate-limit scope.
 * Checked in order — first match wins.
 */
const CODEX_SCOPE_PATTERNS: Array<{ pattern: string; scope: "codex" | "spark" }> = [
  { pattern: "codex-spark", scope: "spark" },
  { pattern: "spark", scope: "spark" },
  { pattern: "codex", scope: "codex" },
  { pattern: "gpt-5", scope: "codex" }, // gpt-5.2-codex, gpt-5.3-codex, etc.
];

/**
 * T09: Determine the rate-limit scope for a Codex model.
 * Use this key as the suffix for per-scope rate limit state:
 *   `${accountId}:${getModelScope(model)}`
 *
 * @param model - The Codex model ID (e.g. "gpt-5.3-codex", "codex-spark-mini")
 * @returns "codex" | "spark"
 */
export function getCodexModelScope(model: string): "codex" | "spark" {
  const lower = model.toLowerCase();
  for (const { pattern, scope } of CODEX_SCOPE_PATTERNS) {
    if (lower.includes(pattern)) return scope;
  }
  return "codex"; // default scope
}

/**
 * T09: Get the scope-keyed rate limit identifier for an account+model combination.
 * Use this as the key for rateLimitState maps to ensure scope isolation.
 */
export function getCodexRateLimitKey(accountId: string, model: string): string {
  return `${accountId}:${getCodexModelScope(model)}`;
}

/**
 * T03: Parsed quota snapshot from Codex response headers.
 * Codex includes per-account usage windows that allow precise reset scheduling.
 * Ref: sub2api PR #357 (feat(oauth): persist usage snapshots and window cooldown)
 */
export interface CodexQuotaSnapshot {
  usage5h: number; // tokens used in 5h window
  limit5h: number; // token limit for 5h window
  resetAt5h: string | null; // ISO timestamp when 5h window resets
  usage7d: number; // tokens used in 7d window
  limit7d: number; // token limit for 7d window
  resetAt7d: string | null; // ISO timestamp when 7d window resets
}

/**
 * T03: Parse Codex-specific quota headers from a provider response.
 * Returns null if none of the relevant headers are present.
 *
 * Extracts:
 *   x-codex-5h-usage / x-codex-5h-limit / x-codex-5h-reset-at
 *   x-codex-7d-usage / x-codex-7d-limit / x-codex-7d-reset-at
 */
export function parseCodexQuotaHeaders(headers: Headers): CodexQuotaSnapshot | null {
  const usage5h = headers.get("x-codex-5h-usage");
  const limit5h = headers.get("x-codex-5h-limit");
  const resetAt5h = headers.get("x-codex-5h-reset-at");
  const usage7d = headers.get("x-codex-7d-usage");
  const limit7d = headers.get("x-codex-7d-limit");
  const resetAt7d = headers.get("x-codex-7d-reset-at");

  // Return null if none of the quota headers are present (not a quota-aware response)
  if (!usage5h && !limit5h && !resetAt5h && !usage7d && !limit7d && !resetAt7d) {
    return null;
  }

  return {
    usage5h: usage5h ? parseFloat(usage5h) : 0,
    limit5h: limit5h ? parseFloat(limit5h) : Infinity,
    resetAt5h: resetAt5h ?? null,
    usage7d: usage7d ? parseFloat(usage7d) : 0,
    limit7d: limit7d ? parseFloat(limit7d) : Infinity,
    resetAt7d: resetAt7d ?? null,
  };
}

/**
 * T03: Get the soonest quota reset time from a CodexQuotaSnapshot.
 * 7d window takes priority (wider window, harder limit) but we use whichever
 * is further in the future to avoid releasing the block too early.
 *
 * @returns Unix timestamp (ms) of the soonest effective reset, or null
 */
export function getCodexResetTime(quota: CodexQuotaSnapshot): number | null {
  const times: number[] = [];
  if (quota.resetAt7d) {
    const t = new Date(quota.resetAt7d).getTime();
    if (!isNaN(t) && t > Date.now()) times.push(t);
  }
  if (quota.resetAt5h) {
    const t = new Date(quota.resetAt5h).getTime();
    if (!isNaN(t) && t > Date.now()) times.push(t);
  }
  if (times.length === 0) return null;
  return Math.max(...times); // Use furthest-out reset to avoid premature unblock
}

/**
 * T03 (Item 3): Compute the minimum-necessary cooldown based on which window
 * is actually exhausted. Prevents over-blocking the account:
 *
 * - If 7d window >= threshold: cooldown until 7d reset (weekly window exhausted)
 * - If 5h window >= threshold: cooldown until 5h reset only (short-term limit)
 * - Otherwise: 0 (account is healthy, no cooldown needed)
 *
 * Called after parsing quota headers from a successful/429 response to
 * mark the account accordingly without overly long cooldowns.
 *
 * @param quota - Parsed quota snapshot from response headers
 * @param threshold - Fraction (0-1) that triggers cooldown (default: 0.95)
 * @returns Cooldown duration in milliseconds (0 = no cooldown needed)
 */
export function getCodexDualWindowCooldownMs(
  quota: CodexQuotaSnapshot,
  threshold = 0.95
): { cooldownMs: number; window: "7d" | "5h" | "none" } {
  const now = Date.now();

  // Compute per-window usage ratios (0..1)
  const ratio7d =
    quota.limit7d > 0 && Number.isFinite(quota.limit7d) ? quota.usage7d / quota.limit7d : 0;
  const ratio5h =
    quota.limit5h > 0 && Number.isFinite(quota.limit5h) ? quota.usage5h / quota.limit5h : 0;

  // 7d window takes priority — if the weekly budget is near-exhausted,
  // we must wait until the weekly reset (not just 5h).
  if (ratio7d >= threshold && quota.resetAt7d) {
    const resetTime = new Date(quota.resetAt7d).getTime();
    if (resetTime > now) {
      return { cooldownMs: resetTime - now, window: "7d" };
    }
  }

  // 5h window (primary short-term rate limit)
  if (ratio5h >= threshold && quota.resetAt5h) {
    const resetTime = new Date(quota.resetAt5h).getTime();
    if (resetTime > now) {
      return { cooldownMs: resetTime - now, window: "5h" };
    }
  }

  return { cooldownMs: 0, window: "none" };
}

// Ordered list of effort levels from lowest to highest
const EFFORT_ORDER = ["none", "low", "medium", "high", "xhigh"] as const;
type EffortLevel = (typeof EFFORT_ORDER)[number];
const CODEX_FAST_WIRE_VALUE = "priority";

function stringifyCodexInstructionContent(content: unknown): string {
  if (typeof content === "string") {
    return content.trim();
  }

  if (Array.isArray(content)) {
    return content
      .map((part) => {
        if (typeof part === "string") return part.trim();
        if (!part || typeof part !== "object") return "";
        const record = part as Record<string, unknown>;
        if (typeof record.text === "string") return record.text.trim();
        if (typeof record.content === "string") return record.content.trim();
        return "";
      })
      .filter(Boolean)
      .join("\n")
      .trim();
  }

  return "";
}

function hoistSystemMessagesToInstructions(body: Record<string, unknown>): void {
  if (!Array.isArray(body.input)) return;

  const systemChunks: string[] = [];
  const filteredInput = body.input.filter((itemValue) => {
    if (!itemValue || typeof itemValue !== "object" || Array.isArray(itemValue)) {
      return true;
    }

    const item = itemValue as Record<string, unknown>;
    const role = typeof item.role === "string" ? item.role : "";
    const type = typeof item.type === "string" ? item.type : "";
    const isSystemMessage = role === "system" && (!type || type === "message");
    if (!isSystemMessage) {
      return true;
    }

    const text = stringifyCodexInstructionContent(item.content);
    if (text) {
      systemChunks.push(text);
    }
    return false;
  });

  if (systemChunks.length === 0) return;

  const existingInstructions =
    typeof body.instructions === "string" ? body.instructions.trim() : "";
  body.instructions = existingInstructions
    ? `${systemChunks.join("\n\n")}\n\n${existingInstructions}`
    : systemChunks.join("\n\n");
  body.input = filteredInput;
}

/**
 * Convert role=system messages in `input` to role=developer.
 *
 * GPT-5 models support the `developer` role in input, but reject `system`.
 * Unlike hoistSystemMessagesToInstructions(), this keeps the content inside
 * the `input` array where it benefits from OpenAI's automatic prompt caching.
 *
 * OpenAI's prompt caching matches on the serialized prefix of the `input` array
 * (+ tools). The `instructions` field is NOT included in the cache key for
 * GPT-5 models. Moving system prompts from `input` to `instructions` therefore
 * removes them from the cacheable prefix, resulting in 0% cache hit rates.
 *
 * Ref: https://community.openai.com/t/caching-is-borked-for-gpt-5-models/1359574
 * Ref: https://community.openai.com/t/no-caching-with-model-responses/1338627
 */
function convertSystemToDeveloperRole(body: Record<string, unknown>): void {
  if (!Array.isArray(body.input)) return;

  for (const itemValue of body.input) {
    if (!itemValue || typeof itemValue !== "object" || Array.isArray(itemValue)) {
      continue;
    }

    const item = itemValue as Record<string, unknown>;
    const role = typeof item.role === "string" ? item.role : "";
    const type = typeof item.type === "string" ? item.type : "";
    const isSystemMessage = role === "system" && (!type || type === "message");
    if (isSystemMessage) {
      item.role = "developer";
    }
  }
}

/**
 * Strip server-generated item IDs from the input array.
 *
 * The Codex /codex/responses endpoint does not persist response items even when
 * store=true is sent. When proxy clients (e.g. OpenClaw) include response items
 * from previous turns in the input array, those items carry server-assigned IDs
 * (prefixed with "rs_", "fc_", "resp_", "msg_"). The Codex backend tries to
 * validate these IDs against its persistence store and returns 404 when the items
 * are not found (because store was effectively false).
 *
 * This function:
 *   1. Removes bare string references ("rs_abc123") from the input array
 *   2. Removes object items with type "item_reference" (explicit stored-item refs)
 *   3. Strips the "id" field from any object in input whose id matches a
 *      server-generated prefix (rs_, fc_, resp_, msg_) — so the content is
 *      preserved but the backend won't try to look it up
 *   4. Always deletes previous_response_id (endpoint doesn't persist responses)
 *
 * @deprecated This function will be removed in v4.0, Codex executor has updated processing logic
 */
function stripStoredItemReferences(body: Record<string, unknown>): void {
  // Always strip previous_response_id — the /codex/responses endpoint does not
  // persist responses, so any reference to a previous response would cause a 404.
  // The official Codex CLI sets previous_response_id to None for HTTP transport.
  // Ref: codex-rs codex-api/src/common.rs:187 — previous_response_id: None
  // Ref: CLIProxyAPI codex_executor.go:115 — sjson.DeleteBytes(body, "previous_response_id")
  delete body.previous_response_id;

  if (!Array.isArray(body.input)) return;

  const SERVER_ID_PATTERN = /^(rs|fc|resp|msg)_/;
  let strippedCount = 0;

  body.input = body.input.filter((item) => {
    // Bare string references: "rs_abc123", "resp_abc123"
    if (typeof item === "string" && SERVER_ID_PATTERN.test(item)) {
      strippedCount++;
      return false;
    }

    // Object references: { type: "item_reference", id: "rs_..." }
    if (
      item &&
      typeof item === "object" &&
      !Array.isArray(item) &&
      (item as Record<string, unknown>).type === "item_reference"
    ) {
      strippedCount++;
      return false;
    }

    // Object items with server-generated IDs: strip the id field but keep the item.
    // e.g. { id: "rs_...", type: "reasoning", summary: [...] } → keep content, remove id
    // e.g. { id: "fc_...", type: "function_call", ... } → keep content, remove id
    if (item && typeof item === "object" && !Array.isArray(item)) {
      const record = item as Record<string, unknown>;
      if (typeof record.id === "string" && SERVER_ID_PATTERN.test(record.id)) {
        delete record.id;
        strippedCount++;
      }
    }

    return true;
  });

  if (strippedCount > 0) {
    console.debug(
      `[Codex] stripStoredItemReferences: sanitized ${strippedCount} server-generated ID(s) from input`
    );
  }
}

function normalizeCodexTools(body: Record<string, unknown>): void {
  if (!Array.isArray(body.tools)) return;

  const validToolNames = new Set<string>();
  body.tools = body.tools.filter((toolValue) => {
    if (!toolValue || typeof toolValue !== "object" || Array.isArray(toolValue)) {
      return false;
    }

    const tool = toolValue as Record<string, unknown>;

    // Preserve namespace tools (MCP tool groups used by Codex/OpenAI Responses API).
    // Codex API supports them natively; register sub-tool names for tool_choice validation.
    if (tool.type === "namespace") {
      if (Array.isArray(tool.tools)) {
        for (const st of tool.tools as unknown[]) {
          if (st && typeof st === "object" && !Array.isArray(st)) {
            const subTool = st as Record<string, unknown>;
            const name = typeof subTool.name === "string" ? subTool.name.trim() : "";
            if (name) validToolNames.add(name);
          }
        }
      }
      return true;
    }

    if (tool.type !== "function") {
      return false;
    }

    const rawName =
      typeof tool.name === "string"
        ? tool.name
        : tool.function &&
            typeof tool.function === "object" &&
            !Array.isArray(tool.function) &&
            typeof (tool.function as Record<string, unknown>).name === "string"
          ? ((tool.function as Record<string, unknown>).name as string)
          : "";
    const name = rawName.trim();
    if (!name) {
      return false;
    }

    validToolNames.add(name);
    return true;
  });

  if (
    body.tool_choice &&
    typeof body.tool_choice === "object" &&
    !Array.isArray(body.tool_choice)
  ) {
    const toolChoice = body.tool_choice as Record<string, unknown>;
    if (toolChoice.type === "function") {
      const rawName = typeof toolChoice.name === "string" ? toolChoice.name.trim() : "";
      if (!rawName || !validToolNames.has(rawName)) {
        delete body.tool_choice;
      }
    }
  }
}

function getResponsesSubpath(endpointPath: unknown): string | null {
  const normalizedEndpoint = String(endpointPath || "").replace(/\/+$/, "");
  const match = normalizedEndpoint.match(/(?:^|\/)responses(?:(\/.*))?$/i);
  if (!match) return null;
  return match[1] || "";
}

function isCompactResponsesEndpoint(endpointPath: unknown): boolean {
  return getResponsesSubpath(endpointPath)?.toLowerCase() === "/compact";
}

function normalizeServiceTierValue(value: unknown): string | undefined {
  if (typeof value !== "string") return undefined;
  const normalized = value.trim().toLowerCase();
  if (!normalized) return undefined;
  if (normalized === "fast") return CODEX_FAST_WIRE_VALUE;
  return normalized;
}

/**
 * Maximum reasoning effort allowed per Codex model.
 * Models not listed here default to "xhigh" (unrestricted).
 * Update this table when Codex releases new models with different caps.
 */
const MAX_EFFORT_BY_MODEL: Record<string, EffortLevel> = {
  "gpt-5.3-codex": "xhigh",
  "gpt-5.2-codex": "xhigh",
  "gpt-5.1-codex-max": "xhigh",
  "gpt-5-mini": "high",
  "gpt-5.1-mini": "high",
  "gpt-4.1-mini": "high",
};

/**
 * Clamp reasoning effort to the model's maximum allowed level.
 * Returns the original value if within limits, or the cap if it exceeds it.
 */
function clampEffort(model: string, requested: string): string {
  const max: EffortLevel = MAX_EFFORT_BY_MODEL[model] ?? "xhigh";
  const reqIdx = EFFORT_ORDER.indexOf(requested as EffortLevel);
  const maxIdx = EFFORT_ORDER.indexOf(max);
  if (reqIdx > maxIdx) {
    console.debug(`[Codex] clampEffort: "${requested}" → "${max}" (model: ${model})`);
    return max;
  }
  return requested;
}

function normalizeEffortValue(value: unknown): string | undefined {
  if (typeof value !== "string") return undefined;
  const normalized = value.trim().toLowerCase();
  return normalized || undefined;
}

function consumeResponsesStoreMarker(body: Record<string, unknown>): unknown {
  const marker = body._omnirouteResponsesStore;
  delete body._omnirouteResponsesStore;
  return marker;
}

/**
 * Codex Executor - handles OpenAI Codex API (Responses API format)
 * Automatically injects default instructions if missing.
 * IMPORTANT: Includes chatgpt-account-id header for workspace binding.
 */
export class CodexExecutor extends BaseExecutor {
  constructor() {
    super("codex", PROVIDERS.codex);
  }

  buildUrl(model, stream, urlIndex = 0, credentials = null) {
    void model;
    void stream;
    void urlIndex;

    const responsesSubpath = getResponsesSubpath(credentials?.requestEndpointPath);
    if (responsesSubpath !== null) {
      const baseUrl = String(this.config.baseUrl || "").replace(/\/$/, "");
      if (baseUrl.endsWith("/responses")) {
        return `${baseUrl}${responsesSubpath}`;
      }
      return `${baseUrl}/responses${responsesSubpath}`;
    }

    return super.buildUrl(model, stream, urlIndex, credentials);
  }

  /**
   * Codex Responses endpoint is SSE-first.
   * Always request event-stream from upstream, even when client requested stream=false.
   * Includes chatgpt-account-id header for strict workspace binding.
   */
  buildHeaders(credentials, stream = true) {
    const isCompactRequest = isCompactResponsesEndpoint(credentials?.requestEndpointPath);
    const headers = super.buildHeaders(credentials, isCompactRequest ? false : true);
    headers.Version = getCodexClientVersion();
    setUserAgentHeader(headers, getCodexUserAgent());

    // Add workspace binding header if workspaceId is persisted
    const workspaceId = credentials?.providerSpecificData?.workspaceId;
    if (workspaceId) {
      headers["chatgpt-account-id"] = workspaceId;
    }

    // Originator header — identifies the client type to the Codex backend.
    // Ref: openai/codex login/src/auth/default_client.rs DEFAULT_ORIGINATOR = "codex_cli_rs"
    headers["originator"] = "codex_cli_rs";

    // session_id header — enables prompt cache affinity on the Codex backend.
    // The official Codex client sets this to conversation_id (a stable UUID per session).
    // Ref: openai/codex codex-api/src/requests/headers.rs build_conversation_headers()
    const cacheSessionId = this.getPromptCacheSessionId(credentials);
    if (cacheSessionId) {
      headers["session_id"] = cacheSessionId;
    }

    return headers;
  }

  /**
   * Derive a stable session ID for prompt cache affinity.
   * Uses workspaceId (chatgpt account ID) as the cache partition key.
   * This mirrors the official Codex client's use of conversation_id for
   * prompt_cache_key and session_id header.
   * Ref: openai/codex core/src/client.rs line 853
   */
  private getPromptCacheSessionId(credentials): string | null {
    return credentials?.providerSpecificData?.workspaceId || null;
  }

  /**
   * Refresh Codex OAuth credentials when a 401 is received.
   * OpenAI uses rotating (one-time-use) refresh tokens — if the token was already
   * consumed by a concurrent refresh, this returns null to signal re-auth is needed.
   *
   * Fixes #251: After a server restart/upgrade, previously cached access tokens may
   * have expired or become invalid. chatCore.ts calls this on 401; previously the
   * base class returned null causing the request to fail instead of refreshing.
   */
  async refreshCredentials(credentials, log) {
    if (!credentials?.refreshToken) {
      log?.warn?.("TOKEN_REFRESH", "Codex: no refresh token available, re-authentication required");
      return null;
    }
    const result = await getAccessToken("codex", credentials, log);
    if (!result || result.error) {
      log?.warn?.(
        "TOKEN_REFRESH",
        `Codex: token refresh failed${result?.error ? ` (${result.error})` : ""} — re-authentication required`
      );
      return null;
    }
    return result;
  }

  /**
   * Transform request before sending - inject default instructions if missing
   */
  transformRequest(model, body, stream, credentials) {
    // Do not mutate the caller's payload in place. Combo quality checks and
    // other post-execute paths still inspect the original request body.
    body =
      body && typeof body === "object" ? structuredClone(body) : ({} as Record<string, unknown>);

    const nativeCodexPassthrough = body?._nativeCodexPassthrough === true;
    const isCompactRequest = isCompactResponsesEndpoint(credentials?.requestEndpointPath);
    const requestDefaults = getCodexRequestDefaults(credentials?.providerSpecificData);
    const thinkingBudgetConfig = getThinkingBudgetConfig();
    const allowConnectionReasoningDefaults = thinkingBudgetConfig.mode === ThinkingMode.PASSTHROUGH;
    consumeResponsesStoreMarker(body);

    // Codex /responses rejects stream=false, but /responses/compact rejects the stream field entirely.
    if (isCompactRequest) {
      delete body.stream;
      delete body.stream_options;
    } else {
      body.stream = true;
    }
    delete body._nativeCodexPassthrough;

    const requestServiceTier = normalizeServiceTierValue(body.service_tier);
    if (requestServiceTier) {
      body.service_tier = requestServiceTier;
    } else if (requestDefaults.serviceTier) {
      body.service_tier = requestDefaults.serviceTier;
    }

    // ── Cache-aware system prompt handling (both paths) ──
    //
    // Convert system → developer role IN-PLACE so system prompts remain in the
    // `input` array where they contribute to the automatic prompt cache prefix.
    // The `instructions` field is NOT included in the cache key for GPT-5 models.
    //
    // This applies to BOTH native passthrough (Responses API) and translated
    // (Chat Completions) paths. Previously the translated path used
    // hoistSystemMessagesToInstructions() which moved system content out of
    // `input` and into `instructions`, destroying cache eligibility.
    //
    // Ref: PR #1346 (original fix for passthrough only)
    convertSystemToDeveloperRole(body);

    if (nativeCodexPassthrough) {
      // Passthrough: minimal placeholder instructions.
      if (
        !body.instructions ||
        (typeof body.instructions === "string" && body.instructions.trim() === "")
      ) {
        body.instructions = "Follow the developer instructions in the conversation.";
      }
    } else {
      // Translated: use CODEX_DEFAULT_INSTRUCTIONS as fallback when no system
      // prompt was provided by the client (safety net for bare requests).
      if (
        !body.instructions ||
        (typeof body.instructions === "string" && body.instructions.trim() === "")
      ) {
        body.instructions = CODEX_DEFAULT_INSTRUCTIONS;
      }
    }

    // Store: The Codex API defaults store to false when not specified.
    // Proxy clients (e.g. OpenClaw) rely on response chaining via previous_response_id,
    // which requires store=true so that response items are persisted.
    // If the client explicitly sets store, respect it. Otherwise default to true.
    const explicitStoreSetting =
      credentials?.providerSpecificData &&
      typeof credentials.providerSpecificData === "object" &&
      !Array.isArray(credentials.providerSpecificData)
        ? credentials.providerSpecificData.openaiStoreEnabled
        : undefined;
    if (explicitStoreSetting === false) {
      body.store = false;
    } else if (body.store === undefined) {
      body.store = true;
    }

    // Codex Responses only supports function tools with non-empty names.
    // Cursor may include custom tools (e.g. ApplyPatch) that work locally but are
    // invalid upstream, and translation bugs can leave orphaned/empty tool_choice names.
    normalizeCodexTools(body);

    // Strip stored response item references (rs_, resp_, msg_ IDs) from input.
    // The /codex/responses endpoint does not persist responses even with store=true,
    // so any references to previous response items would cause 404 errors.
    stripStoredItemReferences(body);

    // Issue #806: Even for native passthrough, some clients (purist completions) might indiscriminately inject
    // a `messages` or `prompt` array which the strict Codex Responses schema rejects.
    delete body.messages;
    delete body.prompt;

    const effortLevels = ["none", "low", "medium", "high", "xhigh"];
    let modelEffort: string | null = null;
    let cleanModel = typeof body.model === "string" ? body.model : model;
    for (const level of effortLevels) {
      if (typeof cleanModel === "string" && cleanModel.endsWith(`-${level}`)) {
        modelEffort = level;
        body.model = cleanModel.slice(0, -`-${level}`.length);
        cleanModel = body.model;
        break;
      }
    }

    const explicitReasoning = normalizeEffortValue(body?.reasoning?.effort);
    const requestReasoningEffort = normalizeEffortValue(body.reasoning_effort);
    const fallbackReasoningEffort = allowConnectionReasoningDefaults
      ? requestDefaults.reasoningEffort || "medium"
      : undefined;
    const rawEffort =
      explicitReasoning || requestReasoningEffort || modelEffort || fallbackReasoningEffort;

    if (explicitReasoning) {
      body.reasoning = {
        ...(body.reasoning && typeof body.reasoning === "object" ? body.reasoning : {}),
        effort: clampEffort(cleanModel, explicitReasoning),
      };
    } else if (rawEffort) {
      body.reasoning = {
        ...(body.reasoning && typeof body.reasoning === "object" ? body.reasoning : {}),
        effort: clampEffort(cleanModel, rawEffort),
      };
    }
    delete body.reasoning_effort;

    // previous_response_id: always stripped by stripStoredItemReferences().
    // The /codex/responses endpoint does not persist responses, so any reference
    // to a previous response ID would cause a 404. This matches the behavior of
    // both the official Codex CLI (sets None) and CLIProxyAPI (deletes the field).

    // Remove unsupported token limit parameters BEFORE the passthrough return.
    // Codex API rejects both max_tokens and max_output_tokens regardless of
    // whether the request came via native passthrough or translation.
    delete body.max_tokens;
    delete body.max_output_tokens;
    delete body.background; // Droid CLI sends this but Codex Responses API rejects it

    // Inject prompt_cache_key for Codex prompt caching.
    // The official Codex client sets this to conversation_id (a stable UUID per session).
    // Ref: openai/codex core/src/client.rs line 853:
    //   let prompt_cache_key = Some(self.client.state.conversation_id.to_string());
    if (!body.prompt_cache_key) {
      const cacheSessionId = this.getPromptCacheSessionId(credentials);
      if (cacheSessionId) {
        body.prompt_cache_key = cacheSessionId;
      }
    }

    // Delete session_id and conversation_id from the body.
    // These are often injected by OmniRoute's fallback logic for store=true,
    // but the upstream Codex API strictly rejects them as unsupported parameters.
    delete body.session_id;
    delete body.conversation_id;

    if (nativeCodexPassthrough) {
      return body;
    }

    // Remove unsupported parameters for Codex API
    delete body.temperature;
    delete body.top_p;
    delete body.frequency_penalty;
    delete body.presence_penalty;
    delete body.logprobs;
    delete body.top_logprobs;
    delete body.n;
    delete body.seed;
    // max_tokens and max_output_tokens already deleted above (before passthrough return)
    delete body.user; // Cursor sends this but Codex doesn't support it
    delete body.prompt_cache_retention; // Cursor sends this but Codex doesn't support it
    delete body.metadata; // Cursor sends this but Codex doesn't support it
    delete body.stream_options; // Cursor sends this but Codex doesn't support it
    delete body.safety_identifier; // Droid CLI sends this but Codex doesn't support it

    return body;
  }
}