@inhuman.tools/harness 0.0.0-stage → 0.2.0

package/dist/providers/openai-compat/actor.js

@@ -0,0 +1,1520 @@
+/**
+ * openai-compat completion actor. The HTTP round is per PEW.
+ *
+ * Connection-wide `#emit` / `#abort` / `#turnModel` would make parallel
+ * remainder and ReACT settle on each other's speech. The connection stays
+ * the durable client: catalog, headers, inflight abort set.
+ *
+ * History, compact lead, frozen/live tools, and squeeze live in harness-context.
+ * viewWire encodes that projection plus vendor reasoning dialect. Chat Completions
+ * vs Responses (`completionBody`) is vendor wire — D4, not context.
+ * System is two messages: constitution+preamble (stable), env/mode (volatile).
+ * Env stays off the first string — a date or mode change must not rewrite the
+ * prefix OpenAI/xAI match from the left.
+ *
+ * Chat Completions vs Responses is a class, chosen in `connection.resolve`
+ * from `completionApi()`. The round does not branch on that token.
+ */
+import { randomUUID } from "node:crypto";
+import { fail, step, trace } from "@inhuman.tools/harness-diagnostics";
+import { assertResponseMatchesSchema, cacheStatusLog, latestCompactCutoff, owningTurn, ResponseCall, JSON_SCHEMA_UNSUPPORTED, SessionRoot, SessionTurn, } from "@inhuman.tools/harness-model";
+import { readBlob, readVisual } from "@inhuman.tools/harness-provider-sdk";
+import { ConnectionMiss, ConnectionNotReady, ExpectationActor, ExpectationInvariant, } from "@inhuman.tools/plexus-expectations/driver";
+import { createStreamSession, mergeReasoning, normalizeMessage, } from "@inhuman.tools/tool-call-normalize";
+import { runInAction } from "mobx";
+import invariant from "tiny-invariant";
+import { approvalNoteMessagesOf as contextApprovalNotes, notificationMessagesOf as contextNotifications, projectHistory, selectCompletionTools, steerMessagesOf as contextSteers, } from "../../context/index.js";
+import { emitsCacheControl, withCacheBreakpoints } from "./anthropic-cache.js";
+import { ARG_ALIASES } from "./arg-aliases.js";
+import { usesExplicitPromptCache } from "./cache-affinity.js";
+import { cacheDumpKind, preferCacheResponse, redactCacheHeaders, writeCacheDump } from "./cache-dump.js";
+import { completionCatalogOf, liveToolDescriptionOf } from "./completion-face.js";
+import { describeHttpFailure, wrapFetchFailure } from "./fetch-failure.js";
+import { parseJsonUnknown, readField } from "./json-read.js";
+import { usageLedger } from "./model-catalog.js";
+import { reasoningDialectFor, } from "./reasoning-flavor.js";
+import { parseStoredToolCall, responseCallsFromToolCalls } from "./tool-call-json.js";
+/** Anthropic/OpenAI prompt-cache default. */
+const PROMPT_CACHE_TTL_MS = 5 * 60 * 1000;
+const COMPLETION_FAILED_STATUS = /(?:openai-compat completion failed: |\bHTTP )(\d{3})\b/;
+const TRANSIENT_CODE = /^(?:ECONNRESET|ETIMEDOUT|ENOTFOUND|ECONNREFUSED|EAI_AGAIN|EPIPE|EHOSTUNREACH|ENETUNREACH|EPROTO|UND_ERR_)/;
+const TRANSIENT_TEXT = /fetch failed|Failed to fetch|ECONNRESET|ETIMEDOUT|ENOTFOUND|ECONNREFUSED|EAI_AGAIN|Connect Timeout|Headers Timeout|other side closed|UND_ERR_|socket hang up/i;
+/** Parsed SSE `data:` only. Socket bytes reset undici's body timer, including pings (`request-timeout.ts`). */
+const STREAM_STALL_MS = 60_000;
+/** Longest wait between attempts, whether it comes from our backoff or from the server. */
+const BACKOFF_CAP_MS = 5 * 60 * 1000;
+/** A 429 or 503 that named how long to wait. The message is the ordinary HTTP failure text. */
+export class RetryAfterError extends Error {
+    waitMs;
+    constructor(message, waitMs) {
+        super(message);
+        this.waitMs = waitMs;
+        this.name = "RetryAfterError";
+    }
+}
+/**
+ * `Retry-After` (seconds or an HTTP date) or the OpenAI-style `retry-after-ms`, read on 429 and
+ * 503 only. Undefined when the response names no usable wait.
+ */
+export function retryAfterMs(headers, status, now = Date.now()) {
+    if (status !== 429 && status !== 503)
+        return undefined;
+    const ms = headers.get("retry-after-ms");
+    if (ms != null && /^\d+(\.\d+)?$/.test(ms.trim()))
+        return Math.round(Number(ms));
+    const raw = headers.get("retry-after")?.trim();
+    if (raw == null || raw === "")
+        return undefined;
+    if (/^\d+(\.\d+)?$/.test(raw))
+        return Math.round(Number(raw) * 1000);
+    const at = Date.parse(raw);
+    return Number.isNaN(at) ? undefined : Math.max(0, at - now);
+}
+class StreamStallError extends Error {
+    constructor(ms) {
+        super(`stream stalled after ${ms}ms`);
+        this.name = "StreamStallError";
+    }
+}
+function isTransientCompletionError(error) {
+    if (error instanceof StreamStallError)
+        return false;
+    if (error instanceof ConnectionMiss)
+        return false;
+    if (error instanceof ExpectationInvariant)
+        return false;
+    if (error instanceof TypeError)
+        return false;
+    const message = error instanceof Error ? error.message : String(error);
+    if (message.includes(JSON_SCHEMA_UNSUPPORTED))
+        return false;
+    const statusMatch = COMPLETION_FAILED_STATUS.exec(message);
+    if (statusMatch) {
+        const status = Number(statusMatch[1]);
+        return status === 408 || status === 429 || status >= 500;
+    }
+    const seen = new Set();
+    let node = error;
+    for (let depth = 0; depth < 6 && node != null; depth += 1) {
+        if (seen.has(node))
+            break;
+        seen.add(node);
+        if (node instanceof TypeError)
+            return true;
+        if (typeof node !== "object") {
+            if (TRANSIENT_TEXT.test(String(node)))
+                return true;
+            break;
+        }
+        const rec = node;
+        if (typeof rec.code === "string" && TRANSIENT_CODE.test(rec.code))
+            return true;
+        if (typeof rec.message === "string" && TRANSIENT_TEXT.test(rec.message))
+            return true;
+        if (rec.name === "TimeoutError" || rec.name === "ConnectTimeoutError" || rec.name === "HeadersTimeoutError") {
+            return true;
+        }
+        node = rec.cause;
+    }
+    return message.includes("openai-compat completion failed");
+}
+function thoughtEcho(thought, field, echo) {
+    if (echo === "omit-empty" && thought === "")
+        return {};
+    return field === "reasoning" ? { reasoning: thought } : { reasoning_content: thought };
+}
+export function chatContentText(content) {
+    if (typeof content === "string")
+        return content;
+    if (content == null)
+        return "";
+    const lines = [];
+    for (const part of content) {
+        if (part.type === "text")
+            lines.push(part.text);
+    }
+    return lines.join("\n");
+}
+function wireContent(message) {
+    const content = message.content;
+    if (typeof content === "string" || content == null)
+        return content;
+    if (message.role !== "user") {
+        const texts = [];
+        for (const part of content) {
+            if (part.type === "text")
+                texts.push({ type: "text", text: part.text });
+        }
+        return texts;
+    }
+    const parts = [];
+    let image = false;
+    for (const part of content) {
+        if (part.type === "text") {
+            parts.push({ type: "text", text: part.text });
+            continue;
+        }
+        if (part.bytes == null)
+            continue;
+        image = true;
+        parts.push({
+            type: "image_url",
+            image_url: { url: `data:${part.mime};base64,${Buffer.from(part.bytes).toString("base64")}` },
+        });
+    }
+    if (!image)
+        return chatContentText(parts);
+    return parts;
+}
+function responsesUserContent(content) {
+    if (typeof content === "string")
+        return content;
+    if (content == null)
+        return "";
+    const parts = [];
+    let image = false;
+    for (const part of content) {
+        if (part.type === "text") {
+            if (part.text !== "")
+                parts.push({ type: "input_text", text: part.text });
+            continue;
+        }
+        image = true;
+        parts.push({ type: "input_image", image_url: part.image_url.url });
+    }
+    if (!image)
+        return chatContentText(content);
+    return parts;
+}
+function asChat(message) {
+    return {
+        role: message.role,
+        content: wireContent(message),
+        ...(message.toolCallId === undefined ? {} : { tool_call_id: message.toolCallId }),
+    };
+}
+export function notificationMessagesOf(expectation) {
+    return contextNotifications(expectation).map(asChat);
+}
+export function approvalNoteMessagesOf(expectation) {
+    return contextApprovalNotes(expectation).map(asChat);
+}
+export function steerMessagesOf(expectation) {
+    return contextSteers(expectation).map(asChat);
+}
+function toOpenAiTool(tool) {
+    return {
+        type: "function",
+        function: {
+            name: tool.name,
+            description: tool.description,
+            parameters: tool.parameters,
+        },
+    };
+}
+/** Catalog tools for this completion's wire. Frozen turn set wins so MCP churn cannot rewrite the prefix. */
+export function toolsFor(expectation) {
+    const root = expectation.plexus?.root;
+    const live = completionCatalogOf(root);
+    const tools = selectCompletionTools(expectation, live);
+    const turn = owningTurn(expectation);
+    if (live == null || (turn != null && turn.tools.entries.length > 0)) {
+        return tools.map(toOpenAiTool);
+    }
+    return tools.map((tool) => {
+        const description = liveToolDescriptionOf(root, tool.name, live);
+        return description === undefined ? toOpenAiTool(tool) : toOpenAiTool({ ...tool, description });
+    });
+}
+/** vLLM-family hosts (Atria) 400 `tools: []`. OpenAI treats the key as optional; absence is "no tools". */
+function wireTools(tools) {
+    return tools.length === 0 ? undefined : { tools };
+}
+/** Hashes already reported missing: every round projects again, and the head fallback is the same each time. */
+const missingBlobsReported = new Set();
+/**
+ * The blob reader the projection gets. A body the document stores beside itself and the bucket cannot give
+ * reaches the model as its head (the projection's rule); this is where the log says which hash was missing.
+ */
+function readBlobReported(sha256) {
+    const bytes = readBlob(sha256);
+    if (bytes === null && !missingBlobsReported.has(sha256)) {
+        missingBlobsReported.add(sha256);
+        fail(`blob ${sha256} is missing from the bucket: the prompt carries its head`);
+    }
+    return bytes;
+}
+export function* viewWire(tip, options = {}) {
+    const echo = options.dialect?.echo ?? "omit-empty";
+    const field = options.dialect?.chatField ?? "reasoning_content";
+    const catalog = completionCatalogOf(tip.plexus?.root);
+    const root = tip.plexus?.root;
+    const sessionId = root instanceof SessionRoot ? root.sessionId : "";
+    for (const message of projectHistory(tip, {
+        catalog,
+        echoEmptyThought: echo === "always",
+        readBlob: readBlobReported,
+        ...(sessionId === "" ? {} : { readVisual: (id) => readVisual(sessionId, id) }),
+    })) {
+        yield encodeChat(message, field, echo);
+    }
+}
+function encodeChat(message, field, echo) {
+    const toolCalls = message.toolCalls == null || message.toolCalls.length === 0
+        ? undefined
+        : message.toolCalls.map((call) => ({
+            id: call.id,
+            type: "function",
+            function: { name: call.name, arguments: call.argsJson },
+        }));
+    const content = wireContent(message);
+    return {
+        role: message.role,
+        // Strict providers reject `content: null` unless the message carries tool_calls.
+        content: content == null && toolCalls === undefined ? "" : content,
+        ...(message.toolCallId === undefined ? {} : { tool_call_id: message.toolCallId }),
+        ...(toolCalls === undefined ? {} : { tool_calls: toolCalls }),
+        ...(message.role === "assistant" ? thoughtEcho(message.thought ?? "", field, echo) : {}),
+        ...(message.thoughtEncrypted == null || message.thoughtEncrypted === ""
+            ? {}
+            : { reasoning_encrypted: message.thoughtEncrypted }),
+    };
+}
+/** Drop reasons that mean "the model meant to call this and the call is unusable". */
+const SURFACED_DROPS = new Set(["truncated", "unparseable_args", "irreparable_args"]);
+/**
+ * Split whole calls from cut-off ones. A call the repair pass closed (`partial`) in a response
+ * that stopped on `length` ended mid-arguments, so it is as untrustworthy as a dropped one.
+ */
+function settleCalls(calls, partial, dropped, finishReason) {
+    const whole = [];
+    const cut = [...dropped];
+    for (const call of calls) {
+        if (finishReason === "length" && (partial.has(call.id) || !parsesAsJson(call.function.arguments))) {
+            cut.push({ id: call.id, name: call.function.name, reason: "truncated" });
+        }
+        else {
+            whole.push(call);
+        }
+    }
+    return { calls: whole, dropped: cut };
+}
+function parsesAsJson(text) {
+    if (text.trim() === "")
+        return true;
+    try {
+        JSON.parse(text);
+        return true;
+    }
+    catch {
+        return false;
+    }
+}
+function responseCallsOf(answer) {
+    return [
+        ...responseCallsFromToolCalls(answer.calls),
+        ...(answer.dropped ?? []).map((drop) => new ResponseCall({ toolCallId: drop.id, name: drop.name, argsJson: "{}", dropReason: drop.reason })),
+    ];
+}
+function reasoningPartText(parts) {
+    if (!Array.isArray(parts))
+        return "";
+    const texts = [];
+    for (const part of parts) {
+        if (typeof part !== "object" || part === null)
+            continue;
+        const text = readField(part, "text");
+        if (typeof text === "string")
+            texts.push(text);
+    }
+    return texts.join("\n\n");
+}
+function visibleThought(raw, summary) {
+    if (raw !== "")
+        return { thought: raw, kind: "raw" };
+    if (summary !== "")
+        return { thought: summary, kind: "summary" };
+    return { thought: "", kind: null };
+}
+/**
+ * Responses `incomplete` + `max_output_tokens` after a reasoning item is the
+ * model spending the budget on thought and stopping before a message. Not a
+ * failed generation — ReACT continues. Other incomplete reasons stay fatal.
+ */
+function incompleteThoughtExhausted(reason, thought) {
+    if (reason !== "max_output_tokens")
+        return false;
+    return (thought.raw.trim() !== "" ||
+        thought.summary.trim() !== "" ||
+        thought.encrypted !== "" ||
+        thought.kind != null ||
+        (thought.reasoningTokens ?? 0) > 0);
+}
+const THINKING_PART_TYPES = new Set(["thinking", "reasoning", "redacted_thinking", "reasoning_text"]);
+function isThinkingPart(part) {
+    const type = readField(part, "type");
+    if (typeof type === "string" && THINKING_PART_TYPES.has(type))
+        return true;
+    return readField(part, "thought") === true;
+}
+function partThought(part) {
+    for (const key of ["thinking", "text", "reasoning", "content"]) {
+        const value = readField(part, key);
+        if (typeof value === "string" && value !== "")
+            return value;
+    }
+    return "";
+}
+/** Speech vs a thinking content part. A thinking part is not the answer. */
+function splitMessageContent(value) {
+    if (typeof value === "string")
+        return { text: value, thought: "", sawThinking: false };
+    if (!Array.isArray(value))
+        return { text: "", thought: "", sawThinking: false };
+    const text = [];
+    const thought = [];
+    let sawThinking = false;
+    for (const part of value) {
+        if (typeof part === "string") {
+            text.push(part);
+            continue;
+        }
+        if (typeof part !== "object" || part === null)
+            continue;
+        if (isThinkingPart(part)) {
+            sawThinking = true;
+            const inner = partThought(part);
+            if (inner !== "")
+                thought.push(inner);
+            continue;
+        }
+        const spoken = readField(part, "text");
+        if (typeof spoken === "string")
+            text.push(spoken);
+    }
+    return { text: text.join(""), thought: thought.join("\n\n"), sawThinking };
+}
+function argAliasesForNormalize() {
+    const out = {};
+    for (const [tool, aliases] of Object.entries(ARG_ALIASES)) {
+        out[tool] = { ...aliases };
+    }
+    return out;
+}
+function catalogToolDefs(expectation) {
+    return toolsFor(expectation).map((tool) => ({
+        name: tool.function.name,
+        parameters: tool.function.parameters,
+    }));
+}
+/** Final non-delta `message.reasoning_content` — stream ingest only reads `delta`. */
+function chunkMessageReasoning(raw) {
+    if (typeof raw !== "object" || raw === null)
+        return "";
+    const choices = readField(raw, "choices");
+    const first = Array.isArray(choices) ? choices[0] : undefined;
+    if (typeof first !== "object" || first === null)
+        return "";
+    const message = readField(first, "message");
+    if (typeof message !== "object" || message === null)
+        return "";
+    const content = readField(message, "reasoning_content");
+    if (typeof content === "string" && content !== "")
+        return content;
+    const alt = readField(message, "reasoning");
+    return typeof alt === "string" && alt !== "" ? alt : "";
+}
+export async function* readSseJson(body, signal, stallMs) {
+    if (!body)
+        return;
+    const reader = body.getReader();
+    const decoder = new TextDecoder();
+    let buf = "";
+    let armTimer;
+    let stalled;
+    const clearArm = () => {
+        if (armTimer === undefined)
+            return;
+        clearTimeout(armTimer);
+        armTimer = undefined;
+    };
+    // Armed when the body opens. A comment or ping is not progress, and neither is a socket that never sends `data:`.
+    const arm = () => {
+        clearArm();
+        const timer = setTimeout(() => {
+            stalled = new StreamStallError(stallMs);
+            void reader.cancel(stalled).catch(() => undefined);
+        }, stallMs);
+        timer.unref();
+        armTimer = timer;
+    };
+    const read = async () => {
+        try {
+            const next = await reader.read();
+            if (stalled !== undefined)
+                throw stalled;
+            return next;
+        }
+        catch (error) {
+            if (stalled !== undefined)
+                throw stalled;
+            throw error;
+        }
+    };
+    try {
+        arm();
+        while (!signal.aborted) {
+            const { done, value } = await read();
+            if (done)
+                break;
+            buf += decoder.decode(value, { stream: true });
+            let yielded = false;
+            for (;;) {
+                const sep = buf.indexOf("\n\n");
+                if (sep === -1)
+                    break;
+                const event = buf.slice(0, sep);
+                buf = buf.slice(sep + 2);
+                for (const line of event.split(/\r?\n/)) {
+                    const trimmed = line.trimEnd();
+                    if (!trimmed.startsWith("data:"))
+                        continue;
+                    const data = trimmed.slice(5).trimStart();
+                    if (!data || data === "[DONE]") {
+                        if (data === "[DONE]")
+                            return;
+                        continue;
+                    }
+                    try {
+                        yield parseJsonUnknown(data);
+                        yielded = true;
+                    }
+                    catch {
+                        // skip malformed line
+                    }
+                }
+            }
+            if (yielded)
+                arm();
+        }
+        if (buf.trim().length > 0) {
+            for (const line of buf.split(/\r?\n/)) {
+                const trimmed = line.trimEnd();
+                if (!trimmed.startsWith("data:"))
+                    continue;
+                const data = trimmed.slice(5).trimStart();
+                if (!data || data === "[DONE]")
+                    continue;
+                try {
+                    yield parseJsonUnknown(data);
+                }
+                catch {
+                    /* skip */
+                }
+            }
+        }
+    }
+    finally {
+        clearArm();
+        try {
+            reader.releaseLock();
+        }
+        catch {
+            /* cancel already dropped the lock */
+        }
+    }
+}
+export class OpenAiCompatActor extends ExpectationActor {
+    driverConnection;
+    #startedAt = 0;
+    #backoffAbort;
+    #requestDump = null;
+    #lastRaw = null;
+    #answerDump = null;
+    /** Same formula as LicenseScope.backoffMs. Tests override. */
+    backoffMs = function (n) {
+        return Math.min(BACKOFF_CAP_MS, Math.max(1000, n * Math.log(n) * 1000));
+    };
+    stallMs = STREAM_STALL_MS;
+    #spokenText = "";
+    #spokenThought = "";
+    constructor(expectation, driverConnection) {
+        super(expectation, driverConnection);
+        this.driverConnection = driverConnection;
+    }
+    reasoningDialect() {
+        return reasoningDialectFor(this.driverConnection.reasoningFlavor(), this.expectation.reasoningEffort, {
+            api: this.driverConnection.completionApi(),
+            model: this.effectiveModel(),
+            hasTools: toolsFor(this.expectation).length > 0,
+        });
+    }
+    /** Abort the inter-attempt sleep. No-op if not sleeping. Does not fail the PEW. */
+    skipBackoff() {
+        this.#backoffAbort?.abort();
+    }
+    async run() {
+        ConnectionNotReady.invariant(this.driverConnection.alive, this.driverConnection.connectionId);
+        if (this.signal.aborted)
+            return;
+        const abort = new AbortController();
+        this.driverConnection.trackInflight(abort);
+        const onActorAbort = () => {
+            abort.abort();
+        };
+        if (this.signal.aborted) {
+            abort.abort();
+        }
+        else {
+            this.signal.addEventListener("abort", onActorAbort, { once: true });
+        }
+        const limit = this.driverConnection.state.windows[this.effectiveModel()] ?? 0;
+        if (limit > 0) {
+            runInAction(() => {
+                this.expectation.contextLimit = limit;
+            });
+        }
+        const runStarted = Date.now();
+        let retryAfter;
+        try {
+            for (let n = 0;; n += 1) {
+                if (n > 0) {
+                    await this.#sleepBackoff(n, retryAfter);
+                    if (this.signal.aborted)
+                        return;
+                }
+                this.#beginAttempt();
+                try {
+                    const turn = owningTurn(this.expectation);
+                    let turnIndex = 1;
+                    if (turn !== null) {
+                        let hops = 0;
+                        let cell = turn;
+                        const seen = new Set();
+                        while (cell != null && !seen.has(cell)) {
+                            seen.add(cell);
+                            hops += 1;
+                            cell = cell.turnTail instanceof SessionTurn ? cell.turnTail : null;
+                        }
+                        turnIndex = Math.max(1, hops);
+                    }
+                    const affinity = this.driverConnection.cacheAffinity(this.cacheSessionId(), this.effectiveModel(), {
+                        requestId: randomUUID(),
+                        turnIndex,
+                    });
+                    const payload = {
+                        ...this.completionBody([
+                            ...viewWire(this.expectation, {
+                                dialect: this.reasoningDialect(),
+                            }),
+                        ]),
+                        ...affinity.body,
+                    };
+                    const headers = this.driverConnection.stamp(affinity.headers);
+                    const url = this.driverConnection.completionRequestUrl();
+                    this.#requestDump = {
+                        url,
+                        headers: redactCacheHeaders(headers),
+                        body: payload,
+                    };
+                    let response;
+                    const posted = Date.now();
+                    trace("harness:complete", `post ${this.expectation.uuid} connection=${this.driverConnection.connectionId} model=${this.effectiveModel()} attempt=${n}`);
+                    try {
+                        response = await this.driverConnection.authorizedFetch(url, {
+                            method: "POST",
+                            headers: affinity.headers,
+                            body: JSON.stringify(payload),
+                        }, abort.signal);
+                    }
+                    catch (error) {
+                        if (error instanceof ConnectionMiss)
+                            throw error;
+                        throw wrapFetchFailure(error, {
+                            why: "completion",
+                            method: "POST",
+                            url,
+                        });
+                    }
+                    if (!response.ok) {
+                        const errorBody = await response.text();
+                        const detail = errorBody.trim().slice(0, 240);
+                        const message = describeHttpFailure({
+                            why: "completion",
+                            method: "POST",
+                            url,
+                            status: response.status,
+                            detail,
+                        });
+                        const waitMs = retryAfterMs(response.headers, response.status);
+                        if (waitMs !== undefined)
+                            throw new RetryAfterError(message, waitMs);
+                        const transientHttp = response.status === 408 || response.status === 429 || response.status >= 500;
+                        const schema = this.expectation.responseSchema;
+                        const schemaSent = schema != null;
+                        if (!transientHttp && schemaSent && response.status >= 400 && response.status < 500) {
+                            throw new Error(`${JSON_SCHEMA_UNSUPPORTED}: ${message}`);
+                        }
+                        throw new Error(message);
+                    }
+                    const answer = await this.readAnswer(response, abort.signal);
+                    trace("harness:complete", `answer ${this.expectation.uuid} ms=${Date.now() - posted} status=${response.status} chars=${answer?.content.length ?? 0}`);
+                    invariant(answer !== null, "openai-compat completion failed");
+                    this.expectation.stampAnswer({
+                        calls: responseCallsOf(answer),
+                        ...(answer.finishReason === undefined ? {} : { finishReason: answer.finishReason }),
+                        text: answer.content,
+                        ...(typeof answer.reasoning === "string" ? { thought: answer.reasoning } : {}),
+                        ...(typeof answer.thoughtKind === "string" ? { thoughtKind: answer.thoughtKind } : {}),
+                    });
+                    this.#answerDump = {
+                        content: answer.content,
+                        reasoning: answer.reasoning,
+                        calls: answer.calls.length,
+                    };
+                    this.#stampWarm();
+                    this.#noteCacheInvariant();
+                    this.speech(answer.content, answer.reasoning);
+                    const schema = this.expectation.responseSchema;
+                    if (schema != null)
+                        assertResponseMatchesSchema(answer.content, JSON.stringify(schema));
+                    if (this.signal.aborted)
+                        return;
+                    this.complete({
+                        text: answer.content,
+                        thought: answer.reasoning,
+                        ...(answer.thoughtKind != null ? { thoughtKind: answer.thoughtKind } : {}),
+                    });
+                    return;
+                }
+                catch (error) {
+                    if (this.signal.aborted)
+                        return;
+                    if (error instanceof ExpectationInvariant)
+                        throw error;
+                    if (!isTransientCompletionError(error)) {
+                        const reason = error instanceof Error ? error.message : String(error);
+                        this.expectation.failReason = reason;
+                        runInAction(() => {
+                            if (this.#spokenText !== "" || this.#spokenThought !== "") {
+                                this.expectation.responseText = this.#spokenText;
+                                this.expectation.responseThought = this.#spokenThought;
+                            }
+                            this.expectation.markThoughtEnded();
+                        });
+                        this.fail(reason);
+                        return;
+                    }
+                    const reason = error instanceof Error ? error.message : String(error);
+                    const exhausted = this.#retryExhausted(n + 1, runStarted);
+                    if (exhausted !== "") {
+                        const final = `${exhausted}: ${reason}`;
+                        this.expectation.failReason = final;
+                        runInAction(() => {
+                            if (this.#spokenText !== "" || this.#spokenThought !== "") {
+                                this.expectation.responseText = this.#spokenText;
+                                this.expectation.responseThought = this.#spokenThought;
+                            }
+                            this.expectation.markThoughtEnded();
+                        });
+                        this.fail(final);
+                        return;
+                    }
+                    retryAfter = error instanceof RetryAfterError ? error.waitMs : undefined;
+                    step(`completion retry ${n + 1}: ${reason}`);
+                }
+            }
+        }
+        finally {
+            this.signal.removeEventListener("abort", onActorAbort);
+            this.driverConnection.untrackInflight(abort);
+            this.#backoffAbort?.abort();
+            this.#backoffAbort = undefined;
+        }
+    }
+    #beginAttempt() {
+        this.#startedAt = Date.now();
+        this.#requestDump = null;
+        this.#lastRaw = null;
+        this.#answerDump = null;
+        this.#spokenText = "";
+        this.#spokenThought = "";
+        this.expectation.beginAttempt(this.driverConnection.cacheFlavor() !== "none" && this.expectation.expectPromptCacheHit());
+        this.report({
+            text: "",
+            thought: "",
+            startedAt: this.#startedAt,
+            cacheHit: this.expectation.cacheHit,
+            expectedCacheHit: this.expectation.expectedCacheHit,
+        });
+    }
+    /**
+     * Empty while the round may retry. The budget is recorded on the expectation when the round
+     * starts; zero is no limit, which is the default.
+     */
+    #retryExhausted(attempts, startedAt) {
+        const { retryMaxAttempts: maxAttempts, retryMaxMs: maxMs } = this.expectation;
+        if (maxAttempts > 0 && attempts >= maxAttempts)
+            return `gave up after ${attempts} attempts`;
+        if (maxMs > 0 && Date.now() - startedAt >= maxMs)
+            return `gave up after ${Math.round((Date.now() - startedAt) / 1000)}s of retrying`;
+        return "";
+    }
+    #sleepBackoff(n, retryAfter) {
+        // Retry-After is a floor on the wait. The cap still bounds it.
+        const delay = Math.max(this.backoffMs(n), Math.min(retryAfter ?? 0, BACKOFF_CAP_MS));
+        if (delay <= 0)
+            return Promise.resolve();
+        const backoff = new AbortController();
+        this.#backoffAbort = backoff;
+        const retryUntilMs = Date.now() + delay;
+        this.report({
+            text: "",
+            thought: "",
+            startedAt: this.#startedAt,
+            retryUntilMs,
+        });
+        return new Promise((resolve) => {
+            let timer;
+            const finish = () => {
+                if (timer !== undefined)
+                    clearTimeout(timer);
+                backoff.signal.removeEventListener("abort", finish);
+                this.signal.removeEventListener("abort", finish);
+                if (this.#backoffAbort === backoff)
+                    this.#backoffAbort = undefined;
+                resolve();
+            };
+            if (backoff.signal.aborted || this.signal.aborted) {
+                finish();
+                return;
+            }
+            timer = setTimeout(finish, delay);
+            backoff.signal.addEventListener("abort", finish, { once: true });
+            this.signal.addEventListener("abort", finish, { once: true });
+        });
+    }
+    speech(text, thought) {
+        if (this.signal.aborted)
+            return;
+        if (text.length === 0 && thought.length === 0)
+            return;
+        this.#spokenText = text;
+        this.#spokenThought = thought;
+        if (text.trim() !== "") {
+            runInAction(() => {
+                this.expectation.markThoughtEnded();
+            });
+        }
+        this.report({
+            text,
+            thought,
+            thoughtKind: this.expectation.responseThoughtKind ?? undefined,
+            startedAt: this.#startedAt,
+            cacheHit: this.expectation.cacheHit,
+            expectedCacheHit: this.expectation.expectedCacheHit,
+        });
+    }
+    stampThoughtKind(kind) {
+        if (kind == null)
+            return;
+        const current = this.expectation.responseThoughtKind;
+        if (current === "raw" || current === kind)
+            return;
+        runInAction(() => {
+            this.expectation.responseThoughtKind = kind;
+        });
+    }
+    #noteCacheInvariant() {
+        const last = this.expectation.report;
+        this.report({
+            text: last?.text ?? "",
+            thought: last?.thought ?? "",
+            thoughtKind: last?.thoughtKind ?? this.expectation.responseThoughtKind ?? undefined,
+            startedAt: this.#startedAt,
+            cacheHit: this.expectation.cacheHit,
+            expectedCacheHit: this.expectation.expectedCacheHit,
+        });
+        const dumped = this.#dumpCacheRound();
+        const note = cacheStatusLog(this.expectation);
+        step(dumped == null ? `completion ${note}` : `completion ${note} dumped ${dumped}`);
+    }
+    #dumpCacheRound() {
+        const root = this.expectation.plexus?.root;
+        const workspacePath = root instanceof SessionRoot ? root.path : "";
+        return writeCacheDump({
+            workspacePath,
+            sessionId: this.cacheSessionId(),
+            startedAt: this.#startedAt,
+            kind: cacheDumpKind(this.expectation.expectedCacheHit, this.expectation.cacheHit),
+            expectedCacheHit: this.expectation.expectedCacheHit,
+            cacheHit: this.expectation.cacheHit,
+            cachedTokens: this.expectation.cachedTokens,
+            totalTokens: this.expectation.totalTokens,
+            request: this.#requestDump,
+            response: this.#lastRaw,
+            answer: this.#answerDump ?? undefined,
+        });
+    }
+    /**
+     * Live-tape ReACT keeps the session id. Remainder / compact / other
+     * off-spine PEWs share `${session}:off-spine` so they cluster with each
+     * other and do not occupy the tape replica — a different prefix on the
+     * same `x-grok-conv-id` evicts the tape KV (xAI cache is per-server).
+     */
+    cacheSessionId() {
+        const root = this.expectation.plexus?.root;
+        const sessionId = root instanceof SessionRoot ? root.sessionId.trim() : "";
+        const base = sessionId === "" ? this.driverConnection.connectionId : sessionId;
+        if (owningTurn(this.expectation) != null)
+            return base;
+        return `${base}:off-spine`;
+    }
+    effectiveModel() {
+        const turn = this.expectation.model.trim();
+        if (turn)
+            return turn;
+        const configured = (this.driverConnection.settings.model ?? "").trim();
+        if (configured)
+            return configured;
+        const live = this.driverConnection.state.models.find((id) => id.trim() !== "")?.trim();
+        invariant(live, "openai-compat model is required");
+        return live;
+    }
+    jsonSchema() {
+        return this.expectation.responseSchema ?? undefined;
+    }
+    stampConsumed(raw) {
+        const ledger = usageLedger(raw);
+        this.#lastRaw = preferCacheResponse(this.#lastRaw, raw);
+        runInAction(() => {
+            if (ledger.totalTokens != null)
+                this.expectation.totalTokens = ledger.totalTokens;
+            if (ledger.cachedTokens != null) {
+                this.expectation.cachedTokens = ledger.cachedTokens;
+                this.expectation.cacheHit = ledger.cachedTokens > 0;
+            }
+            if (ledger.inputTokens != null)
+                this.expectation.inputTokens = ledger.inputTokens;
+            if (ledger.outputTokens != null)
+                this.expectation.outputTokens = ledger.outputTokens;
+            if (ledger.cacheWriteTokens != null)
+                this.expectation.cacheWriteTokens = ledger.cacheWriteTokens;
+            if (ledger.reasoningTokens != null)
+                this.expectation.reasoningTokens = ledger.reasoningTokens;
+            if (ledger.costUsd != null)
+                this.expectation.costUsd = ledger.costUsd;
+        });
+    }
+    #stampWarm() {
+        runInAction(() => {
+            this.expectation.warmUntil = new Date(Date.now() + PROMPT_CACHE_TTL_MS);
+        });
+    }
+}
+export class OpenAiCompatChatActor extends OpenAiCompatActor {
+    /** Anthropic `cache_control` breakpoints, where the route takes them and the connection allows. */
+    cacheMarked(messages, model) {
+        const settings = this.driverConnection.settings;
+        if (!emitsCacheControl(settings, this.driverConnection.cacheFlavor(), model))
+            return messages;
+        const turn = owningTurn(this.expectation);
+        const card = turn == null ? null : latestCompactCutoff(turn);
+        return withCacheBreakpoints(messages, card?.job.standIn ?? "", settings.cacheControlTtl ?? "5m");
+    }
+    completionBody(messages) {
+        const schema = this.jsonSchema();
+        const model = this.effectiveModel();
+        const openaiExplicit = this.driverConnection.cacheFlavor() === "openai" && usesExplicitPromptCache(model);
+        const first = messages[0];
+        // GPT-5.6+ implicit mode writes through the latest eligible message. An explicit
+        // breakpoint on the stable system head lets a later env/mode rewrite still reuse
+        // constitution+tools. `instructions` cannot hold a breakpoint — Chat Completions
+        // system parts can.
+        // https://developers.openai.com/api/docs/guides/prompt-caching
+        const stamped = openaiExplicit && first != null && first.role === "system" && typeof first.content === "string"
+            ? [
+                {
+                    ...first,
+                    content: [
+                        {
+                            type: "text",
+                            text: first.content,
+                            prompt_cache_breakpoint: { mode: "explicit" },
+                        },
+                    ],
+                },
+                ...messages.slice(1),
+            ]
+            : messages;
+        return {
+            model,
+            messages: this.cacheMarked(stamped, model),
+            stream: true,
+            stream_options: { include_usage: true },
+            temperature: this.driverConnection.settings.temperature,
+            ...wireTools(toolsFor(this.expectation)),
+            ...this.reasoningDialect().body,
+            ...(schema === undefined
+                ? {}
+                : {
+                    response_format: {
+                        type: "json_schema",
+                        json_schema: { name: "Output", schema, strict: true },
+                    },
+                }),
+        };
+    }
+    async readAnswer(response, signal) {
+        const contentType = response.headers.get("content-type")?.toLowerCase() ?? "";
+        if (contentType.includes("application/json")) {
+            return this.#roundFromBulk(await response.json());
+        }
+        if (contentType.includes("text/event-stream") || contentType.includes("event-stream")) {
+            return this.#consumeStream(readSseJson(response.body, signal, this.stallMs));
+        }
+        const text = await response.text();
+        const head = text.trimStart();
+        if (head.startsWith("{") || head.startsWith("[")) {
+            try {
+                const parsed = JSON.parse(text);
+                return this.#roundFromBulk(parsed);
+            }
+            catch {
+                return null;
+            }
+        }
+        return this.#consumeStream(readSseJson(new ReadableStream({
+            start(controller) {
+                controller.enqueue(new TextEncoder().encode(text));
+                controller.close();
+            },
+        }), signal, this.stallMs));
+    }
+    #roundFromBulk(payload) {
+        this.stampConsumed(payload);
+        const choices = typeof payload === "object" && payload !== null ? readField(payload, "choices") : undefined;
+        const choice = Array.isArray(choices) && typeof choices[0] === "object" && choices[0] !== null ? choices[0] : undefined;
+        const message = choice == null ? undefined : readField(choice, "message");
+        const messageRec = typeof message === "object" && message !== null ? message : undefined;
+        const reasoningContent = messageRec == null ? undefined : readField(messageRec, "reasoning_content");
+        const reasoningAlt = messageRec == null ? undefined : readField(messageRec, "reasoning");
+        const reasoning = typeof reasoningContent === "string" && reasoningContent.length > 0
+            ? reasoningContent
+            : typeof reasoningAlt === "string" && reasoningAlt.length > 0
+                ? reasoningAlt
+                : "";
+        const content = messageRec == null ? undefined : readField(messageRec, "content");
+        const spoken = splitMessageContent(content);
+        const toolCalls = messageRec == null ? undefined : readField(messageRec, "tool_calls");
+        const finishReason = choice == null ? undefined : readField(choice, "finish_reason");
+        const normalized = normalizeMessage({
+            content: spoken.text,
+            tool_calls: Array.isArray(toolCalls)
+                ? toolCalls.flatMap((item) => {
+                    const call = parseStoredToolCall(item);
+                    return call == null ? [] : [call];
+                })
+                : null,
+            reasoning_content: mergeReasoning(reasoning || null, spoken.thought),
+        }, {
+            mode: "repair",
+            tools: catalogToolDefs(this.expectation),
+            argAliases: argAliasesForNormalize(),
+            modelHint: this.effectiveModel(),
+            finishReasonIn: typeof finishReason === "string" ? finishReason : null,
+        });
+        const reasoningText = normalized.reasoning_content ?? "";
+        const thoughtKind = reasoningText !== "" || spoken.sawThinking ? "raw" : null;
+        this.stampThoughtKind(thoughtKind);
+        const answer = {
+            content: normalized.content ?? "",
+            reasoning: reasoningText,
+            thoughtKind,
+            calls: normalized.tool_calls,
+            finishReason: normalized.finish_reason ?? undefined,
+            dropped: normalized.meta.dropped
+                .filter((drop) => SURFACED_DROPS.has(drop.reason) && drop.name !== "")
+                .map((drop, index) => ({ id: `dropped_${index}`, name: drop.name, reason: drop.reason })),
+        };
+        this.speech(answer.content, answer.reasoning);
+        return answer;
+    }
+    async #consumeStream(chunks) {
+        const session = createStreamSession({
+            mode: "repair",
+            tools: catalogToolDefs(this.expectation),
+            argAliases: argAliasesForNormalize(),
+            modelHint: this.effectiveModel(),
+        });
+        let thoughtAcc = "";
+        let textAcc = "";
+        let sawThinking = false;
+        const calls = [];
+        const partial = new Set();
+        const dropped = [];
+        let finishReason;
+        const onEvent = (ev) => {
+            switch (ev.type) {
+                case "reasoning_delta": {
+                    if (!ev.text)
+                        break;
+                    if (thoughtAcc === "" && textAcc === "") {
+                        trace("harness:complete", `first-token ${this.expectation.uuid} kind=thought`);
+                    }
+                    thoughtAcc += ev.text;
+                    this.stampThoughtKind("raw");
+                    this.speech(textAcc, thoughtAcc);
+                    break;
+                }
+                case "text_delta": {
+                    if (!ev.text)
+                        break;
+                    if (textAcc === "" && thoughtAcc === "") {
+                        trace("harness:complete", `first-token ${this.expectation.uuid} kind=text`);
+                    }
+                    textAcc += ev.text;
+                    this.speech(textAcc, thoughtAcc);
+                    break;
+                }
+                case "tool_call_end": {
+                    if (!ev.name || ev.arguments == null || !ev.id)
+                        break;
+                    if (ev.partial === true)
+                        partial.add(ev.id);
+                    calls.push({
+                        id: ev.id,
+                        type: "function",
+                        function: { name: ev.name, arguments: ev.arguments },
+                    });
+                    break;
+                }
+                case "tool_call_drop": {
+                    if (ev.id && ev.name && ev.reason !== undefined && SURFACED_DROPS.has(ev.reason)) {
+                        dropped.push({ id: ev.id, name: ev.name, reason: ev.reason });
+                    }
+                    break;
+                }
+                case "finish": {
+                    if (ev.finish_reason !== undefined)
+                        finishReason = ev.finish_reason;
+                    break;
+                }
+                default:
+                    break;
+            }
+        };
+        for await (const raw of chunks) {
+            this.stampConsumed(raw);
+            if (typeof raw === "object" && raw !== null) {
+                const rec = raw;
+                const choice = rec.choices?.[0];
+                const delta = choice?.delta;
+                let chunk = rec;
+                if (delta != null) {
+                    const altRaw = readField(delta, "reasoning");
+                    const alt = typeof altRaw === "string" ? altRaw : "";
+                    const rawContent = readField(delta, "content");
+                    const parts = Array.isArray(rawContent) ? splitMessageContent(rawContent) : null;
+                    if (parts?.sawThinking === true)
+                        sawThinking = true;
+                    const hasReasoning = typeof delta.reasoning_content === "string" && delta.reasoning_content.length > 0;
+                    const reasoning = hasReasoning
+                        ? `${delta.reasoning_content}${parts?.thought ?? ""}`
+                        : `${alt}${parts?.thought ?? ""}`;
+                    if (parts != null || (alt.length > 0 && !hasReasoning)) {
+                        chunk = {
+                            ...rec,
+                            choices: [
+                                {
+                                    ...choice,
+                                    delta: {
+                                        ...delta,
+                                        ...(parts == null ? {} : { content: parts.text }),
+                                        ...(reasoning.length > 0 ? { reasoning_content: reasoning } : {}),
+                                    },
+                                },
+                                ...(rec.choices?.slice(1) ?? []),
+                            ],
+                        };
+                    }
+                }
+                for (const ev of session.ingest(chunk))
+                    onEvent(ev);
+            }
+            if (thoughtAcc === "")
+                thoughtAcc = chunkMessageReasoning(raw);
+        }
+        for (const ev of session.close())
+            onEvent(ev);
+        const settled = settleCalls(calls, partial, dropped, finishReason);
+        return {
+            content: textAcc,
+            reasoning: thoughtAcc,
+            thoughtKind: thoughtAcc !== "" || sawThinking ? "raw" : null,
+            calls: settled.calls,
+            dropped: settled.dropped,
+            ...(finishReason === undefined ? {} : { finishReason }),
+        };
+    }
+}
+export class OpenAiCompatResponsesActor extends OpenAiCompatActor {
+    completionBody(messages) {
+        const schema = this.jsonSchema();
+        const isolated = this.expectation.completionMode === "isolated";
+        const firstSystem = messages.find((message) => message.role === "system");
+        const dialect = this.reasoningDialect();
+        const reasoning = dialect.body.reasoning;
+        const input = [];
+        if (isolated) {
+            for (const { role, content } of messages)
+                input.push({ role, content });
+        }
+        else {
+            let skippedFirstSystem = false;
+            for (const message of messages) {
+                if (message.role === "system") {
+                    if (!skippedFirstSystem) {
+                        skippedFirstSystem = true;
+                        continue;
+                    }
+                    input.push({
+                        role: "developer",
+                        content: chatContentText(message.content),
+                    });
+                    continue;
+                }
+                if (message.role === "tool") {
+                    input.push({
+                        type: "function_call_output",
+                        call_id: message.tool_call_id,
+                        output: chatContentText(message.content),
+                    });
+                    continue;
+                }
+                if (message.role === "assistant") {
+                    const encrypted = message.reasoning_encrypted?.trim() ?? "";
+                    const thought = (message.reasoning_content ?? message.reasoning ?? "").trim();
+                    if (encrypted !== "") {
+                        input.push({
+                            type: "reasoning",
+                            // Codex 400s without this key. Empty is valid; thought becomes summary_text.
+                            summary: thought === "" ? [] : [{ type: "summary_text", text: thought }],
+                            encrypted_content: encrypted,
+                        });
+                    }
+                    const calls = message.tool_calls ?? [];
+                    const text = chatContentText(message.content);
+                    if (text !== "") {
+                        input.push({ role: "assistant", content: text });
+                    }
+                    else if (calls.length === 0 && encrypted === "") {
+                        input.push({ role: "assistant", content: "" });
+                    }
+                    for (const call of calls) {
+                        input.push({
+                            type: "function_call",
+                            call_id: call.id,
+                            name: call.function.name,
+                            arguments: call.function.arguments,
+                        });
+                    }
+                    continue;
+                }
+                input.push({ role: "user", content: responsesUserContent(message.content) });
+            }
+        }
+        return {
+            model: this.effectiveModel(),
+            instructions: isolated ? "" : typeof firstSystem?.content === "string" ? firstSystem.content : undefined,
+            input,
+            ...wireTools(toolsFor(this.expectation).map((tool) => ({
+                type: "function",
+                name: tool.function.name,
+                description: tool.function.description,
+                parameters: tool.function.parameters,
+            }))),
+            stream: true,
+            // ChatGPT Codex 400s store:true. Encrypted reasoning is echoed instead.
+            store: false,
+            include: ["reasoning.encrypted_content"],
+            // Visible thought is the summary, not the encrypted CoT. Codex CLI sends this.
+            reasoning: typeof reasoning === "object" && reasoning !== null ? { summary: "auto", ...reasoning } : { summary: "auto" },
+            // Responses structured output is text.format, not response_format.
+            ...(schema === undefined
+                ? {}
+                : {
+                    text: {
+                        format: {
+                            type: "json_schema",
+                            name: "Output",
+                            strict: true,
+                            schema,
+                        },
+                    },
+                }),
+        };
+    }
+    async readAnswer(response, signal) {
+        const contentType = response.headers.get("content-type") ?? "";
+        if (contentType.toLowerCase().includes("application/json") && !contentType.toLowerCase().includes("event-stream")) {
+            const payload = await response.json();
+            this.stampConsumed(payload);
+            const parts = [];
+            const rawParts = [];
+            const summaryParts = [];
+            const calls = [];
+            let encrypted = "";
+            let sawReasoning = false;
+            const collectItem = (item) => {
+                if (typeof item !== "object" || item === null)
+                    return;
+                const type = readField(item, "type");
+                const callId = readField(item, "call_id");
+                const name = readField(item, "name");
+                const args = readField(item, "arguments");
+                if (type === "function_call" &&
+                    typeof callId === "string" &&
+                    typeof name === "string" &&
+                    typeof args === "string") {
+                    calls.push({
+                        id: callId,
+                        type: "function",
+                        function: { name, arguments: args },
+                    });
+                    return;
+                }
+                if (type === "reasoning") {
+                    sawReasoning = true;
+                    const encryptedContent = readField(item, "encrypted_content");
+                    if (typeof encryptedContent === "string" && encryptedContent !== "") {
+                        encrypted = encryptedContent;
+                    }
+                    const raw = reasoningPartText(readField(item, "content"));
+                    const summary = reasoningPartText(readField(item, "summary"));
+                    if (raw !== "")
+                        rawParts.push(raw);
+                    if (summary !== "")
+                        summaryParts.push(summary);
+                }
+                const content = readField(item, "content");
+                if (type === "message" && Array.isArray(content)) {
+                    for (const part of content) {
+                        if (typeof part !== "object" || part === null)
+                            continue;
+                        const text = readField(part, "text");
+                        if (readField(part, "type") === "output_text" && typeof text === "string")
+                            parts.push(text);
+                    }
+                }
+            };
+            const output = typeof payload === "object" && payload !== null ? readField(payload, "output") : undefined;
+            if (Array.isArray(output)) {
+                for (const item of output)
+                    collectItem(item);
+            }
+            const text = parts.join("");
+            const visible = visibleThought(rawParts.join("\n\n"), summaryParts.join("\n\n"));
+            const thoughtKind = visible.kind ?? (sawReasoning || encrypted !== "" ? "raw" : null);
+            this.#stampReasoningEncrypted(encrypted);
+            this.stampThoughtKind(thoughtKind);
+            this.speech(text, visible.thought);
+            return { content: text, reasoning: visible.thought, thoughtKind, calls };
+        }
+        let acc = {
+            text: "",
+            raw: "",
+            summary: "",
+            lastSummaryPart: null,
+        };
+        const calls = [];
+        let encrypted = "";
+        let sawTerminal = false;
+        let finish;
+        const partial = new Set();
+        const emitSpeech = () => {
+            const visible = visibleThought(acc.raw, acc.summary);
+            this.stampThoughtKind(visible.kind);
+            this.speech(acc.text, visible.thought);
+        };
+        for await (const raw of readSseJson(response.body, signal, this.stallMs)) {
+            this.stampConsumed(raw);
+            if (typeof raw !== "object" || raw === null)
+                continue;
+            const type = readField(raw, "type");
+            const delta = readField(raw, "delta");
+            switch (type) {
+                case "error":
+                case "response.failed": {
+                    const eventError = readField(raw, "error");
+                    const eventResponse = readField(raw, "response");
+                    const responseError = typeof eventResponse === "object" && eventResponse !== null ? readField(eventResponse, "error") : undefined;
+                    const eventMessage = typeof eventError === "object" && eventError !== null ? readField(eventError, "message") : undefined;
+                    const nestedMessage = typeof responseError === "object" && responseError !== null
+                        ? readField(responseError, "message")
+                        : undefined;
+                    const detail = eventMessage ?? nestedMessage;
+                    throw new Error(typeof detail === "string" ? detail : `Responses stream failed: ${String(type)}`);
+                }
+                case "response.completed":
+                case "response.done": {
+                    sawTerminal = true;
+                    break;
+                }
+                case "response.incomplete": {
+                    const eventResponse = readField(raw, "response");
+                    const incomplete = typeof eventResponse === "object" && eventResponse !== null
+                        ? readField(eventResponse, "incomplete_details")
+                        : undefined;
+                    const details = typeof incomplete === "object" && incomplete !== null ? readField(incomplete, "reason") : undefined;
+                    const nestedOutput = typeof eventResponse === "object" && eventResponse !== null
+                        ? readField(eventResponse, "output")
+                        : undefined;
+                    if (Array.isArray(nestedOutput)) {
+                        for (const item of nestedOutput) {
+                            if (typeof item !== "object" || item === null)
+                                continue;
+                            const itemType = readField(item, "type");
+                            const callId = readField(item, "call_id");
+                            const name = readField(item, "name");
+                            const args = readField(item, "arguments");
+                            if (itemType === "function_call" &&
+                                typeof callId === "string" &&
+                                typeof name === "string" &&
+                                typeof args === "string" &&
+                                !calls.some((call) => call.id === callId)) {
+                                partial.add(callId);
+                                calls.push({
+                                    id: callId,
+                                    type: "function",
+                                    function: { name, arguments: args },
+                                });
+                                continue;
+                            }
+                            if (itemType === "reasoning") {
+                                const encryptedContent = readField(item, "encrypted_content");
+                                if (encrypted === "" && typeof encryptedContent === "string" && encryptedContent !== "") {
+                                    encrypted = encryptedContent;
+                                }
+                                const thoughtRaw = reasoningPartText(readField(item, "content"));
+                                const summary = reasoningPartText(readField(item, "summary"));
+                                if (acc.raw === "" && thoughtRaw !== "")
+                                    acc = { ...acc, raw: thoughtRaw };
+                                if (acc.summary === "" && summary !== "")
+                                    acc = { ...acc, summary };
+                                this.stampThoughtKind("raw");
+                                continue;
+                            }
+                            if (itemType === "message" && acc.text === "") {
+                                const content = readField(item, "content");
+                                if (!Array.isArray(content))
+                                    continue;
+                                const texts = [];
+                                for (const part of content) {
+                                    if (typeof part !== "object" || part === null)
+                                        continue;
+                                    const text = readField(part, "text");
+                                    if (readField(part, "type") === "output_text" && typeof text === "string")
+                                        texts.push(text);
+                                }
+                                if (texts.length > 0)
+                                    acc = { ...acc, text: texts.join("") };
+                            }
+                        }
+                    }
+                    if (incompleteThoughtExhausted(details, {
+                        raw: acc.raw,
+                        summary: acc.summary,
+                        encrypted,
+                        kind: this.expectation.responseThoughtKind,
+                        reasoningTokens: this.expectation.reasoningTokens,
+                    })) {
+                        sawTerminal = true;
+                        finish = "length";
+                        break;
+                    }
+                    // Speech cut off by max_output_tokens stays a failed round (chatgpt-complete.test); only
+                    // the filter is a stop of its own.
+                    if (details === "content_filter") {
+                        sawTerminal = true;
+                        finish = "content_filter";
+                        break;
+                    }
+                    throw new Error(`Responses stream incomplete${typeof details === "string" ? `: ${details}` : ""}`);
+                }
+                case "response.output_item.done": {
+                    const item = readField(raw, "item");
+                    if (typeof item === "object" && item !== null) {
+                        const itemType = readField(item, "type");
+                        const callId = readField(item, "call_id");
+                        const name = readField(item, "name");
+                        const args = readField(item, "arguments");
+                        const encryptedContent = readField(item, "encrypted_content");
+                        if (itemType === "function_call" &&
+                            typeof callId === "string" &&
+                            typeof name === "string" &&
+                            typeof args === "string") {
+                            calls.push({
+                                id: callId,
+                                type: "function",
+                                function: { name, arguments: args },
+                            });
+                        }
+                        else if (itemType === "reasoning") {
+                            if (typeof encryptedContent === "string" && encryptedContent !== "") {
+                                encrypted = encryptedContent;
+                            }
+                            this.stampThoughtKind("raw");
+                        }
+                    }
+                    break;
+                }
+                default:
+                    if (type === "response.output_text.delta" && typeof delta === "string") {
+                        acc = { ...acc, text: acc.text + delta };
+                        emitSpeech();
+                    }
+                    else if (type === "response.reasoning_summary_text.delta" && typeof delta === "string") {
+                        const indexRaw = readField(raw, "summary_index");
+                        const itemIdRaw = readField(raw, "item_id");
+                        const index = typeof indexRaw === "number" ? indexRaw : 0;
+                        const itemId = typeof itemIdRaw === "string" ? itemIdRaw : "";
+                        const partKey = `${itemId}:${index}`;
+                        const gap = acc.lastSummaryPart !== null && partKey !== acc.lastSummaryPart && acc.summary !== "" ? "\n\n" : "";
+                        acc = {
+                            ...acc,
+                            summary: acc.summary + gap + delta,
+                            lastSummaryPart: partKey,
+                        };
+                        emitSpeech();
+                    }
+                    else if (type === "response.reasoning_text.delta" && typeof delta === "string") {
+                        acc = { ...acc, raw: acc.raw + delta };
+                        emitSpeech();
+                    }
+            }
+        }
+        if (!sawTerminal)
+            throw new Error("Responses stream ended before terminal response event");
+        this.#stampReasoningEncrypted(encrypted);
+        const visible = visibleThought(acc.raw, acc.summary);
+        const thoughtKind = visible.kind ?? (encrypted !== "" || this.expectation.responseThoughtKind != null ? "raw" : null);
+        const settled = settleCalls(calls, partial, [], finish);
+        return {
+            content: acc.text,
+            reasoning: visible.thought,
+            thoughtKind,
+            calls: settled.calls,
+            dropped: settled.dropped,
+            finishReason: finish ?? (calls.length > 0 ? "tool_calls" : "stop"),
+        };
+    }
+    #stampReasoningEncrypted(encrypted) {
+        if (encrypted === "")
+            return;
+        runInAction(() => {
+            this.expectation.responseReasoningEncrypted = encrypted;
+        });
+    }
+}
+//# sourceMappingURL=actor.js.map