kimi-plugin-cc 2.4.5 → 2.5.2

package/plugins/kimi-dsh/dist/cli-client.js

@@ -0,0 +1,876 @@
+// Subprocess client for `kimi -p --output-format stream-json`.
+//
+// Per-job env (`KIMI_PLUGIN_CC_CMD`) is set on this spawn so the
+// PreToolUse hook can branch per command (review vs rescue vs ask). It is
+// passed via the spawn `env` option, not exported to a shell, so concurrent
+// jobs never share an env block.
+import { spawn } from "node:child_process";
+import { appendFile, mkdir } from "node:fs/promises";
+import path from "node:path";
+import { assertPromptFitsCommandLine } from "./prompt-size.js";
+import { createHookEscapeMonitor } from "./hook-escape.js";
+import { RuntimeError, formatError } from "./errors.js";
+import { assertNoUnsafeExperimentalSelector, assertExecutionPlanMatchesSpawn, } from "./kimi-engine.js";
+import { assertPrefixArgsSafe } from "./kimi-command.js";
+import { resolveKimiHome } from "./kimi-home.js";
+import { assertNativeV2Preflight } from "./native-v2-preflight.js";
+import { parseVersionLine } from "./kimi-version-probe.js";
+import { StreamJsonParser, extractSessionIdFromStderr, } from "./stream-json.js";
+import { collectDescendantIdentities, revalidateProcessIdentities, waitForProcessTreeExit, } from "./process-tree.js";
+/** Bytes of stderr retained for diagnostics on completion. Rolling buffer. */
+const STDERR_TAIL_BYTES = 8192;
+/** Hard cap on awaiting diagnostics log drain before resolving the result. */
+const LOG_DRAIN_TIMEOUT_MS = 250;
+/** Default 1500 ms delay between SIGTERM and SIGKILL after an abort. */
+const DEFAULT_ESCALATION_MS = 1_500;
+/** Maximum time spent confirming POSIX tree exit after SIGKILL dispatch. */
+const POST_KILL_QUIESCENCE_MS = 500;
+/** Hard bound for the one-shot descendant identity snapshot. */
+const DESCENDANT_SNAPSHOT_TIMEOUT_MS = 1_000;
+/** Hard bound for each batched identity revalidation pass. */
+const IDENTITY_REVALIDATION_TIMEOUT_MS = 250;
+const DESCENDANT_SNAPSHOT_TIMEOUT = Symbol("descendant-snapshot-timeout");
+function spawnErrorDetails(error) {
+    const cause = error;
+    return {
+        spawn_code: cause?.code,
+        spawn_errno: cause?.errno,
+        spawn_message: formatError(error),
+    };
+}
+export async function runCliPrompt(opts) {
+    // Pre-aborted signal: refuse to spawn at all.
+    if (opts.signal?.aborted === true) {
+        throw new RuntimeError("CLI_ABORTED", "kimi subprocess request cancelled before spawn", "cli-client.pre-spawn", { details: { command: opts.command } });
+    }
+    assertExecutionPlanMatchesSpawn(opts.executionPlan, opts.command, opts.prefixArgs ?? [], opts.env);
+    if ((opts.commandLabel === "rescue" || opts.commandLabel === "swarm-write") &&
+        !opts.trustedWorkspaceRoot) {
+        throw new RuntimeError("CLI_TRUSTED_WORKSPACE_ROOT_MISSING", `Refusing ${opts.commandLabel}: write-capable plugin launches require a plugin-owned trusted workspace root.`, "cli-client.pre-spawn", { details: { command_label: opts.commandLabel } });
+    }
+    // The master experimental flag enables v2-only features (tower, subagent
+    // fork) outside the certified no-plan profile. Refuse it at the final
+    // boundary too, in case the environment changed after plan creation.
+    assertNoUnsafeExperimentalSelector(opts.env, "cli-client.pre-spawn");
+    // Validate the ACTUAL prefix argv used for this spawn — not only the
+    // env-resolved value — so a forged/corrupt persisted plan cannot smuggle a
+    // reserved kimi flag (e.g. `-r <session>`, which would resume a session the
+    // v2 preflight never scanned for plan taint).
+    assertPrefixArgsSafe(opts.prefixArgs ?? []);
+    if (opts.executionPlan.intendedEngine === "native-v2") {
+        // Re-run the cheap no-plan preconditions at the final boundary: the
+        // operator's config.toml and the resume journal are mutable between plan
+        // creation (possibly in another process) and this spawn.
+        await assertNativeV2Preflight({
+            kimiHome: resolveKimiHome(opts.env, opts.cwd),
+            env: opts.env,
+            resumeSessionId: opts.resumeSessionId,
+            stage: "cli-client.spawn",
+        });
+    }
+    const args = buildArgs(opts);
+    assertPromptFitsCommandLine(opts.command, args);
+    const env = buildKimiCliEnvironment(opts);
+    if (opts.logPath !== undefined) {
+        await mkdir(path.dirname(opts.logPath), { recursive: true }).catch(() => {
+            // Best-effort; if mkdir fails the appendFile below will error and be
+            // swallowed by the log chain (diagnostics are non-blocking).
+        });
+    }
+    let child;
+    let spawnedAt;
+    try {
+        spawnedAt = Date.now();
+        child = spawn(opts.command, args, {
+            cwd: opts.cwd,
+            env,
+            stdio: "pipe",
+            // POSIX gets a fresh process group so cancellation reaches kimi's
+            // descendants; Windows has no negative-pid process-group signaling.
+            detached: process.platform !== "win32",
+        });
+    }
+    catch (err) {
+        throw new RuntimeError("CLI_SPAWN_FAILED", `Failed to spawn kimi: ${formatError(err)}`, "cli-client.spawn", err instanceof Error
+            ? { cause: err, details: { command: opts.command, args, ...spawnErrorDetails(err) } }
+            : { details: { command: opts.command, args, ...spawnErrorDetails(err) } });
+    }
+    // Pre-attach an 'error' listener immediately. Bun (and Node) can emit
+    // 'error' asynchronously between spawn returning and our Promise body
+    // attaching the real listener — without a handler in place, EventEmitter
+    // crashes the process. Capture into an array (not a single slot) so two
+    // back-to-back errors don't lose the first; we replay the first one when
+    // the real handler installs. The replacement is atomic across two
+    // synchronous statements (flag-flip + .on attach), so no async work can
+    // interleave between them.
+    const earlyErrors = [];
+    let earlyErrorHandled = false;
+    child.on("error", function earlyErrorListener(err) {
+        if (earlyErrorHandled)
+            return;
+        earlyErrors.push(err);
+    });
+    const parser = new StreamJsonParser();
+    const records = [];
+    const malformed = [];
+    // Rolling stderr tail — keep only the trailing STDERR_TAIL_BYTES so a
+    // long-running kimi process emitting megabytes of stderr (tool.progress,
+    // thinking deltas) doesn't grow our RSS unboundedly.
+    let stderrTail = "";
+    let announcedSessionId;
+    let announcedGoalSummary;
+    let announcedCliError;
+    let announcedSystemVersion;
+    const hookMonitor = createHookEscapeMonitor({ commandLabel: opts.commandLabel,
+        operationKind: opts.executionPlan.operationKind, trustedWorkspaceRoot: opts.trustedWorkspaceRoot, cwd: opts.cwd });
+    const hookEscapes = [];
+    const noteHookEscape = (escape) => {
+        hookEscapes.push(escape);
+        appendLogLine({ event: "hook_escape", ...escape, command_label: opts.commandLabel,
+            operation_kind: opts.executionPlan.operationKind });
+    };
+    let engineMismatchDetected = false;
+    let handleEngineMismatch;
+    const plannedEngine = opts.executionPlan.intendedEngine;
+    const plannedVersion = opts.executionPlan.kimiVersion === null
+        ? null
+        : (parseVersionLine(opts.executionPlan.kimiVersion)?.raw ?? opts.executionPlan.kimiVersion);
+    let logChain = Promise.resolve();
+    const appendLogLine = (payload) => {
+        if (opts.logPath === undefined)
+            return;
+        const line = `${JSON.stringify({ ts: new Date().toISOString(), ...payload })}\n`;
+        logChain = logChain.then(() => appendFile(opts.logPath, line, "utf8").catch(() => {
+            // Diagnostics best-effort; never crash on log failure.
+        }));
+    };
+    appendLogLine({
+        event: "spawn",
+        command: opts.command,
+        args,
+        cwd: opts.cwd,
+        command_label: opts.commandLabel ?? null,
+        operation_kind: opts.executionPlan.operationKind,
+        intended_engine: opts.executionPlan.intendedEngine,
+        kimi_version: opts.executionPlan.kimiVersion,
+        plan_certification: opts.executionPlan.certification,
+        safety_profile: opts.executionPlan.safetyProfile,
+        swarm_max_concurrency: opts.swarmMaxConcurrency ?? null,
+        legacy_v1_forced: env.KIMI_CODE_LEGACY_FLAG === KIMI_LEGACY_FORCED_VALUE,
+    });
+    const invokeOnRecord = (record) => {
+        if (opts.onRecord === undefined)
+            return;
+        try {
+            opts.onRecord(record);
+        }
+        catch {
+            // Caller-supplied callback must never destabilize the parse loop.
+        }
+    };
+    // A native-v2 plan must see agent-core-v2's pre-tool system.version marker
+    // before any consumer-facing output; a legacy plan must never see it.
+    const requireMarkerBeforeOutput = (what) => {
+        if (plannedEngine !== "native-v2" || announcedSystemVersion !== undefined)
+            return false;
+        handleEngineMismatch?.({
+            observedEngine: null,
+            systemVersion: undefined,
+            reason: `${what} arrived before the native-v2 system.version marker; the engine that started cannot be established.`,
+        });
+        return true;
+    };
+    const consumeOutcomes = (outcomes) => {
+        for (const outcome of outcomes) {
+            if (engineMismatchDetected)
+                return;
+            if (outcome.cliError !== undefined) {
+                if (requireMarkerBeforeOutput("a CLI error record"))
+                    return;
+                announcedCliError ??= outcome.cliError;
+                appendLogLine({ event: "cli_error", ...outcome.cliError });
+                continue;
+            }
+            if (outcome.unknownRecord !== undefined) {
+                // Unknown roles are diagnostic-only and excluded from consumer records;
+                // native-v2 still refuses any output before its marker.
+                if (requireMarkerBeforeOutput(`an unknown-role (${outcome.unknownRecord.role}) record`)) {
+                    return;
+                }
+                appendLogLine({
+                    event: "unknown_record",
+                    role: outcome.unknownRecord.role,
+                });
+                continue;
+            }
+            if (outcome.goalSummary !== undefined) {
+                if (requireMarkerBeforeOutput("goal.summary"))
+                    return;
+                // Out-of-band metadata; first-announce-wins, excluded from consumer records.
+                if (announcedGoalSummary === undefined) {
+                    announcedGoalSummary = outcome.goalSummary;
+                    appendLogLine({
+                        event: "goal_summary",
+                        goal_id: outcome.goalSummary.goalId,
+                        status: outcome.goalSummary.status,
+                        turns_used: outcome.goalSummary.turnsUsed,
+                        tokens_used: outcome.goalSummary.tokensUsed,
+                        wall_clock_ms: outcome.goalSummary.wallClockMs,
+                    });
+                }
+                continue;
+            }
+            if (outcome.record !== undefined) {
+                // Meta records are out-of-band wrapper metadata, never consumer-facing
+                // assistant/tool records. Capture session.resume_hint (first-announce
+                // wins, mirroring the stderr capture below for kimi 0.1.x), and skip
+                // every recognized meta type, including 0.23.5's turn.step.retrying and
+                // agent-core-v2's system.version. This keeps records[] and onRecord
+                // stable while preserving the resume hint as the primary session-id
+                // source under kimi 0.2.0+.
+                if (outcome.record.role === "meta") {
+                    // Contract: on a native-v2 plan the system.version marker is the
+                    // FIRST stream-json line. Any other meta (e.g. session.resume_hint)
+                    // arriving before it violates the ordering and means the started
+                    // engine cannot be established — refuse before pinning a session id.
+                    if (outcome.record.type !== "system.version" &&
+                        requireMarkerBeforeOutput(`a ${outcome.record.type} record`)) {
+                        return;
+                    }
+                    if (outcome.record.type === "session.resume_hint" &&
+                        announcedSessionId === undefined) {
+                        announcedSessionId = outcome.record.sessionId;
+                        appendLogLine({
+                            event: "session_announce",
+                            source: "stream-json.meta",
+                            session_id: outcome.record.sessionId,
+                        });
+                    }
+                    else if (outcome.record.type === "system.version" &&
+                        announcedSystemVersion === undefined) {
+                        announcedSystemVersion = outcome.record.version;
+                        appendLogLine({
+                            event: "system_version",
+                            source: "stream-json.meta",
+                            version: outcome.record.version,
+                        });
+                        if (plannedEngine !== "native-v2") {
+                            handleEngineMismatch?.({
+                                observedEngine: "native-v2",
+                                systemVersion: outcome.record.version,
+                                reason: `the pre-tool system.version marker proves native-v2 ${outcome.record.version} started.`,
+                            });
+                            return;
+                        }
+                        // The marker's `version` is a structured field, not a CLI banner:
+                        // require a single well-formed semver token. A multi-line or
+                        // otherwise decorated value must not be normalized into a certified
+                        // one by taking its first line.
+                        const observedVersion = parseExactVersionToken(outcome.record.version);
+                        if (observedVersion === undefined) {
+                            handleEngineMismatch?.({
+                                observedEngine: "native-v2",
+                                systemVersion: outcome.record.version,
+                                reason: `the system.version marker carries a malformed version value ${JSON.stringify(outcome.record.version)}; the started engine's version cannot be established.`,
+                            });
+                            return;
+                        }
+                        if (plannedVersion !== null && observedVersion !== plannedVersion) {
+                            handleEngineMismatch?.({
+                                observedEngine: "native-v2",
+                                systemVersion: outcome.record.version,
+                                reason: `the system.version marker reports ${outcome.record.version}, but the plan certified ${plannedVersion}.`,
+                            });
+                            return;
+                        }
+                    }
+                    continue;
+                }
+                if (requireMarkerBeforeOutput(`a ${outcome.record.role} record`))
+                    return;
+                const escape = hookMonitor.observe(outcome.record);
+                if (escape)
+                    noteHookEscape(escape);
+                records.push(outcome.record);
+                appendLogLine({ event: "record", record: outcome.record });
+                invokeOnRecord(outcome.record);
+            }
+            else if (outcome.malformedLine !== undefined) {
+                // Diagnostics-only in general, but on a native-v2 plan a line that is
+                // not the marker arriving first still violates marker-first: the
+                // engine that produced it cannot be established.
+                if (requireMarkerBeforeOutput("a malformed stream-json line"))
+                    return;
+                const entry = {
+                    line: truncateChars(outcome.malformedLine, 200),
+                    reason: outcome.malformedReason ?? "unknown",
+                };
+                malformed.push(entry);
+                appendLogLine({ event: "malformed", ...entry });
+            }
+        }
+    };
+    const errorDetails = () => ({
+        command: opts.command,
+        args,
+        command_label: opts.commandLabel ?? null,
+    });
+    return await new Promise(async (resolve, reject) => {
+        let aborted = false;
+        let cancellationTeardown;
+        let settlementStarted = false;
+        const settle = (kind, payload) => {
+            if (settlementStarted)
+                return;
+            // Denied calls still without a result (a cancel/budget race, an abnormal
+            // exit, or a turn that ended without the denial the hook owes) are reported
+            // alongside the real outcome, never instead of it.
+            for (const escape of hookMonitor.settle())
+                noteHookEscape(escape);
+            if (kind === "resolve" && hookEscapes.length > 0)
+                payload = { ...payload, hookEscapes: [...hookEscapes] };
+            settlementStarted = true;
+            opts.signal?.removeEventListener("abort", onAbort);
+            // Abort teardown can outlive the direct child's close event while a
+            // detached descendant is still being enumerated or escalated. Never
+            // expose a result/error until that one owned teardown path completes.
+            void (cancellationTeardown ?? Promise.resolve({ ok: true })).then((teardown) => {
+                if (!teardown.ok) {
+                    appendLogLine({
+                        event: "cancellation_teardown_error",
+                        message: formatError(teardown.error),
+                    });
+                    reject(new RuntimeError("CLI_CANCELLATION_TEARDOWN_FAILED", `kimi cancellation teardown failed: ${formatError(teardown.error)}`, "cli-client.cancellation", teardown.error instanceof Error
+                        ? { cause: teardown.error, details: errorDetails() }
+                        : { details: errorDetails() }));
+                    return;
+                }
+                if (kind === "resolve")
+                    resolve({ ...payload, settledAt: Date.now() });
+                else
+                    reject(payload);
+            });
+        };
+        const escalationMs = opts.escalationMs ?? DEFAULT_ESCALATION_MS;
+        const identityReader = opts.processIdentityReader;
+        let processExited = false;
+        child.once("exit", () => {
+            processExited = true;
+        });
+        const captureDescendantIdentities = async (childPid) => {
+            let snapshot;
+            try {
+                const collected = await raceWithValueTimeout((opts.descendantIdentityCollector ?? collectDescendantIdentities)(childPid), DESCENDANT_SNAPSHOT_TIMEOUT_MS, DESCENDANT_SNAPSHOT_TIMEOUT);
+                if (collected === DESCENDANT_SNAPSHOT_TIMEOUT) {
+                    appendLogLine({
+                        event: "descendant_collection_timeout",
+                        timeout_ms: DESCENDANT_SNAPSHOT_TIMEOUT_MS,
+                    });
+                    return [];
+                }
+                snapshot = collected;
+            }
+            catch (err) {
+                appendLogLine({
+                    event: "descendant_collection_error",
+                    message: formatError(err),
+                });
+                return [];
+            }
+            const identities = [];
+            const seen = new Set();
+            for (const identity of snapshot) {
+                const key = `${identity.pid}:${identity.startToken}:${identity.processGroupId}`;
+                if (Number.isInteger(identity.pid) &&
+                    identity.pid > 0 &&
+                    identity.pid !== childPid &&
+                    Number.isInteger(identity.processGroupId) &&
+                    identity.processGroupId > 0 &&
+                    identity.startToken.length > 0 &&
+                    !seen.has(key)) {
+                    seen.add(key);
+                    identities.push({ ...identity });
+                }
+            }
+            return identities;
+        };
+        const trySignalPid = (pid, signal) => {
+            try {
+                process.kill(pid, signal);
+            }
+            catch (err) {
+                const ignored = isErrnoException(err, "ESRCH") || isErrnoException(err, "EPERM");
+                void ignored;
+            }
+        };
+        const signalCapturedTree = async (childPid, identities, signal) => {
+            // The live ChildProcess anchors its detached group. Signal the group
+            // before the direct child can exit and make that numeric PGID stale.
+            if (!processExited) {
+                trySignalPid(-childPid, signal);
+                try {
+                    child.kill(signal);
+                }
+                catch {
+                    // Best-effort direct-child signaling; ChildProcess owns this pid.
+                }
+            }
+            // One bounded pass replaces up to 512 sequential ps calls. A pid whose
+            // start token or PGID changed belongs to another process and is skipped.
+            const pidValidation = await revalidateProcessIdentities(identities, {
+                identityReader,
+                timeoutMs: IDENTITY_REVALIDATION_TIMEOUT_MS,
+            });
+            for (const current of pidValidation.identities) {
+                trySignalPid(current.pid, signal);
+            }
+            // Revalidate again immediately before signaling non-root groups. Never
+            // send to an unanchored numeric PGID left over from collection.
+            const groupValidation = await revalidateProcessIdentities(identities, {
+                identityReader,
+                timeoutMs: IDENTITY_REVALIDATION_TIMEOUT_MS,
+            });
+            const activeGroupIds = new Set(groupValidation.identities.map((identity) => identity.processGroupId));
+            activeGroupIds.delete(childPid);
+            for (const processGroupId of activeGroupIds) {
+                trySignalPid(-processGroupId, signal);
+            }
+        };
+        const teardownChildTree = async () => {
+            await opts.cancellationTeardownHook?.();
+            const childPid = child.pid;
+            if (childPid === undefined || processExited)
+                return;
+            const finiteEscalation = Number.isFinite(escalationMs);
+            if (process.platform === "win32") {
+                // Preserve the documented win32 behavior: only the direct child is
+                // managed; descendant enumeration/reaping remains unsupported.
+                try {
+                    child.kill("SIGTERM");
+                }
+                catch {
+                    // Best-effort.
+                }
+                const exited = await waitForProcessTreeExit([], escalationMs, {
+                    isRootAlive: () => !processExited,
+                });
+                if (!exited && finiteEscalation) {
+                    try {
+                        child.kill("SIGKILL");
+                    }
+                    catch {
+                        // Best-effort.
+                    }
+                }
+                return;
+            }
+            const identities = await captureDescendantIdentities(childPid);
+            await signalCapturedTree(childPid, identities, "SIGTERM");
+            // The grace interval starts only after SIGTERM has reached every
+            // revalidated target. Slow process enumeration never consumes the grace.
+            const exited = await waitForProcessTreeExit(identities, escalationMs, {
+                identityReader,
+                isRootAlive: () => !processExited,
+            });
+            if (!exited && finiteEscalation) {
+                await signalCapturedTree(childPid, identities, "SIGKILL");
+                const quiescent = await waitForProcessTreeExit(identities, POST_KILL_QUIESCENCE_MS, {
+                    identityReader,
+                    isRootAlive: () => !processExited,
+                });
+                if (!quiescent) {
+                    appendLogLine({
+                        event: "post_kill_quiescence_timeout",
+                        timeout_ms: POST_KILL_QUIESCENCE_MS,
+                        descendant_count: identities.length,
+                    });
+                }
+            }
+        };
+        const stopForPolicyFailure = (error) => {
+            engineMismatchDetected = true; // Stops all further consumer output.
+            cancellationTeardown ??= teardownChildTree().then(() => ({ ok: true }), (error) => ({ ok: false, error }));
+            settle("reject", error);
+        };
+        let engineMismatchError;
+        handleEngineMismatch = (mismatch) => {
+            if (engineMismatchError !== undefined)
+                return;
+            engineMismatchDetected = true;
+            engineMismatchError = new RuntimeError("CLI_ENGINE_PROVENANCE_MISMATCH", `Refusing kimi output: execution plan required ${plannedEngine}, but ${mismatch.reason}`, "cli-client.engine-provenance", {
+                details: {
+                    operation_kind: opts.executionPlan.operationKind,
+                    intended_engine: plannedEngine,
+                    observed_engine: mismatch.observedEngine,
+                    planned_kimi_version: opts.executionPlan.kimiVersion,
+                    system_version: mismatch.systemVersion ?? null,
+                },
+            });
+            appendLogLine({
+                event: "engine_provenance_mismatch",
+                intended_engine: plannedEngine,
+                observed_engine: mismatch.observedEngine,
+                system_version: mismatch.systemVersion ?? null,
+                reason: mismatch.reason,
+            });
+            stopForPolicyFailure(engineMismatchError);
+        };
+        const onAbort = () => {
+            if (aborted)
+                return;
+            aborted = true;
+            cancellationTeardown = teardownChildTree().then(() => ({ ok: true }), (error) => ({ ok: false, error }));
+            void cancellationTeardown.then((outcome) => {
+                if (outcome.ok)
+                    return;
+                // A bug in teardown must neither strand the outer promise nor leave the
+                // owned root running. The controlled child anchors this emergency kill.
+                const childPid = child.pid;
+                if (childPid !== undefined && !processExited) {
+                    if (process.platform !== "win32")
+                        trySignalPid(-childPid, "SIGKILL");
+                    try {
+                        child.kill("SIGKILL");
+                    }
+                    catch {
+                        // Best-effort; settlement still surfaces the teardown failure.
+                    }
+                }
+                settle("reject", new RuntimeError("CLI_CANCELLATION_TEARDOWN_FAILED", "kimi cancellation teardown failed", "cli-client.cancellation"));
+            });
+        };
+        opts.signal?.addEventListener("abort", onAbort, { once: true });
+        // Abort may occur during any pre-spawn await, and addEventListener does not
+        // replay an already-aborted signal. Re-check after attaching the listener so
+        // the spawned child cannot escape cancellation teardown.
+        if (opts.signal?.aborted === true) {
+            onAbort();
+        }
+        if (child.pid !== undefined && opts.onSpawn) {
+            try {
+                await opts.onSpawn(child.pid);
+            }
+            catch (error) {
+                onAbort();
+                settle("reject", new RuntimeError("CLI_SPAWN_CALLBACK_FAILED", `Failed to record the spawned kimi process: ${formatError(error)}`, "cli-client.spawn", error instanceof Error ? { cause: error, details: errorDetails() } : { details: errorDetails() }));
+                return;
+            }
+        }
+        earlyErrorHandled = true;
+        child.on("error", (err) => {
+            appendLogLine({ event: "process_error", message: formatError(err) });
+            settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess error: ${formatError(err)}`, "cli-client.process", err instanceof Error
+                ? { cause: err, details: { ...errorDetails(), ...spawnErrorDetails(err) } }
+                : { details: { ...errorDetails(), ...spawnErrorDetails(err) } }));
+        });
+        if (earlyErrors.length > 0) {
+            const captured = earlyErrors[0];
+            queueMicrotask(() => {
+                appendLogLine({
+                    event: "process_error",
+                    message: formatError(captured),
+                    early: true,
+                    additional_count: earlyErrors.length - 1,
+                });
+                settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess error: ${formatError(captured)}`, "cli-client.process", { cause: captured, details: { ...errorDetails(), ...spawnErrorDetails(captured) } }));
+            });
+        }
+        child.on("close", async (exitCode, signal) => {
+            consumeOutcomes(parser.flush());
+            if (!engineMismatchDetected &&
+                exitCode === 0 &&
+                !aborted &&
+                requireMarkerBeforeOutput("a clean exit")) {
+                return;
+            }
+            // The stderr resume-hint channel is the kimi 0.1.x fallback and exists
+            // only for legacy-v1 plans; a native-v2 plan pins a session id from the
+            // stdout meta record alone (after the marker), so stderr can never pin
+            // an id ahead of, or different from, the provenance-checked stream.
+            const sessionId = announcedSessionId ??
+                (plannedEngine === "legacy-v1" ? extractSessionIdFromStderr(stderrTail) : undefined);
+            appendLogLine({
+                event: "exit",
+                exit_code: exitCode,
+                signal,
+                session_id: sessionId,
+                malformed_count: malformed.length,
+                record_count: records.length,
+                aborted,
+            });
+            // Cap the log drain wait so a wedged disk can't block result delivery.
+            // The result itself is in memory; diagnostics are best-effort.
+            await raceWithTimeout(logChain, LOG_DRAIN_TIMEOUT_MS);
+            settle("resolve", {
+                spawnedAt,
+                settledAt: 0,
+                sessionId,
+                goalSummary: announcedGoalSummary,
+                cliError: announcedCliError,
+                systemVersion: announcedSystemVersion,
+                observedEngine: announcedSystemVersion !== undefined
+                    ? "native-v2"
+                    : plannedEngine === "legacy-v1" &&
+                        (exitCode === 0 ||
+                            records.length > 0 ||
+                            sessionId !== undefined ||
+                            announcedGoalSummary !== undefined)
+                        ? "legacy-v1"
+                        : null,
+                records,
+                malformed,
+                stderrTail,
+                exitCode: exitCode ?? -1,
+                signal,
+                aborted,
+            });
+        });
+        // Stdin: kimi -p reads from argv, not stdin. End it immediately so the
+        // child never blocks. Stream method calls on a failed-spawn child can
+        // throw synchronously — guard so the 'error' event still wins.
+        try {
+            child.stdout.setEncoding("utf8");
+            child.stderr.setEncoding("utf8");
+            child.stdout.on("data", (chunk) => {
+                consumeOutcomes(parser.push(chunk));
+            });
+            child.stderr.on("data", (chunk) => {
+                // First-announce-wins across both channels: once a session id is
+                // pinned (from the stdout meta record or an earlier stderr line),
+                // never let a later stderr line overwrite it. The stderr channel is
+                // the 0.1.x fallback and is consulted ONLY for legacy-v1 plans — on a
+                // native-v2 plan a stderr line could otherwise pin a session id before
+                // (or different from) the provenance-checked stdout stream.
+                if (plannedEngine === "legacy-v1" && announcedSessionId === undefined) {
+                    const nextSessionId = extractSessionIdFromStderr(stderrTail + chunk);
+                    if (nextSessionId !== undefined) {
+                        announcedSessionId = nextSessionId;
+                    }
+                }
+                stderrTail = appendToTail(stderrTail, chunk, STDERR_TAIL_BYTES);
+            });
+            child.stdin.end();
+        }
+        catch (err) {
+            settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess stream setup failed: ${formatError(err)}`, "cli-client.streams", err instanceof Error
+                ? { cause: err, details: { ...errorDetails(), ...spawnErrorDetails(err) } }
+                : { details: { ...errorDetails(), ...spawnErrorDetails(err) } }));
+        }
+    });
+}
+/**
+ * Helper for callers that require a session id (replay-capable commands).
+ * Throws CLI_NO_SESSION_ID if the result lacks one, embedding the command
+ * label and exit code in details so the failure is actionable in logs.
+ *
+ * Use this AT the call site rather than baking the requirement into
+ * runCliPrompt — some callers (setup probes, smoke tests) are fine with
+ * an undefined sessionId.
+ */
+export function requireSessionId(result, context) {
+    if (result.sessionId !== undefined && result.sessionId.length > 0) {
+        return result.sessionId;
+    }
+    throw new RuntimeError("CLI_NO_SESSION_ID", "kimi subprocess completed without announcing a session id; replay and resume will be unavailable", "cli-client.session-id", {
+        details: {
+            command_label: context.commandLabel ?? null,
+            exit_code: result.exitCode,
+            signal: result.signal,
+            aborted: result.aborted,
+            stderr_tail_len: result.stderrTail.length,
+        },
+    });
+}
+/** Settled timeout data; stream records are deliberately non-serializable.
+ * Without a partial result (teardown rejection), aborted means abort was issued,
+ * not that the process was observed aborting. */
+export class ResponseTimeoutError extends RuntimeError {
+    constructor(budgetMs, stage, commandLabel, partialResult, cause) {
+        super("RESPONSE_TIMEOUT", `${stage} timed out after ${budgetMs}ms.`, stage, {
+            ...(cause instanceof Error ? { cause } : {}),
+            details: {
+                budget_ms: budgetMs,
+                command_label: commandLabel ?? null,
+                record_count: partialResult?.records.length ?? 0,
+                partial_output_chars: partialResult?.records.reduce((count, record) => count + (record.role === "assistant" && typeof record.content === "string" ? record.content.length : 0), 0) ?? 0,
+                aborted: partialResult?.aborted ?? true,
+                ...(partialResult ? { settled_post_abort: true } : {}),
+                ...(partialResult?.hookEscapes?.length ? { hook_escapes: partialResult.hookEscapes } : {}),
+            },
+        });
+        this.name = "ResponseTimeoutError";
+        // Never serialize stream records into job error details or error envelopes.
+        Object.defineProperty(this, "partialResult", { value: partialResult, enumerable: false });
+    }
+}
+/**
+ * Run `kimi -p` under a budget that also tells the subprocess to die
+ * when the budget expires.
+ *
+ * Why this exists separately from `withTimeout`:
+ *
+ *   The generic `withTimeout` (runtime/kimi-timeouts.ts) only rejects
+ *   the outer Promise — it has no handle on the kimi child process.
+ *
+ * Behavior:
+ *
+ *   - Owns an internal AbortController. If the caller passes
+ *     `opts.signal`, both signals are linked — abort on either side
+ *     aborts the internal controller.
+ *   - When the budget expires, abort the internal controller, await the
+ *     subprocess/tree teardown, then reject with
+ *     `RuntimeError("RESPONSE_TIMEOUT", …)` matching the legacy
+ *     `withTimeout` code shape.
+ *   - A clean non-aborted exit-0 result wins same-tick budget expiry.
+ *     Otherwise a settled result on timeout is carried as partial output.
+ */
+export async function runCliPromptWithBudget(opts, budgetMs, stage) {
+    const controller = new AbortController();
+    if (opts.signal?.aborted === true) {
+        controller.abort();
+    }
+    const onParentAbort = () => controller.abort();
+    opts.signal?.addEventListener("abort", onParentAbort, { once: true });
+    let timeoutFired = false;
+    let timer;
+    timer = setTimeout(() => {
+        timeoutFired = true;
+        controller.abort();
+    }, budgetMs);
+    timer.unref();
+    const timeoutError = (partialResult, cause) => new ResponseTimeoutError(budgetMs, stage, opts.commandLabel, partialResult, cause);
+    try {
+        let result;
+        try {
+            // Deliberately do not race this promise against an early timeout
+            // rejection. The timer aborts it, and runCliPrompt owns the complete
+            // child/descendant teardown before it can settle.
+            result = await runCliPrompt({ ...opts, signal: controller.signal });
+        }
+        catch (error) {
+            if (timeoutFired)
+                throw timeoutError(undefined, error);
+            throw error;
+        }
+        if (timeoutFired && (result.aborted || result.exitCode !== 0)) {
+            // A clean, non-aborted result won the settle/timeout boundary and is success.
+            // Budget aborts normally resolve with records, after owned teardown settles.
+            throw timeoutError(result);
+        }
+        return result;
+    }
+    finally {
+        if (timer !== undefined)
+            clearTimeout(timer);
+        opts.signal?.removeEventListener("abort", onParentAbort);
+    }
+}
+/**
+ * Parse a structured version FIELD (the system.version marker), not a CLI
+ * banner: the whole value must be one semver token (optional leading `v`,
+ * optional pre-release/build suffix) with no surrounding whitespace or extra
+ * lines. Returns the normalized `raw` form, or undefined when malformed.
+ */
+function parseExactVersionToken(value) {
+    if (/\s/.test(value) || value.length === 0)
+        return undefined;
+    return parseVersionLine(value)?.raw;
+}
+const KIMI_LEGACY_ENV = "KIMI_CODE_LEGACY_FLAG";
+const KIMI_LEGACY_FORCED_VALUE = "1";
+function buildArgs(opts) {
+    const args = [...(opts.prefixArgs ?? [])];
+    args.push("--output-format", "stream-json");
+    if (opts.resumeSessionId !== undefined) {
+        args.push("-r", opts.resumeSessionId);
+    }
+    if (opts.model !== undefined) {
+        args.push("-m", opts.model);
+    }
+    if (opts.skillsDirs !== undefined) {
+        for (const dir of opts.skillsDirs) {
+            args.push("--skills-dir", dir);
+        }
+    }
+    // -p accepts the prompt as the option value. We pass it as a separate argv
+    // entry so prompts with leading dashes or whitespace are not interpreted as
+    // additional flags.
+    args.push("-p", opts.prompt);
+    return args;
+}
+export function buildKimiCliEnvironment(opts) {
+    const env = { ...opts.env };
+    if (opts.thinkingEffort !== undefined)
+        env.KIMI_MODEL_THINKING_EFFORT = opts.thinkingEffort;
+    // Ambient workspace trust starts project MCP stdio servers outside PreToolUse.
+    delete env.KIMI_CODE_TRUST_WORKSPACE;
+    // Legacy-v1 plans use this flag on releases whose unflagged -p default can
+    // route to native-v2. Always overwrite ambient false values; older releases
+    // ignore it. The engine choice stays plugin-owned, while certified native-v2
+    // plans use the no-plan construction rather than a legacy pin.
+    if (opts.executionPlan.intendedEngine === "legacy-v1") {
+        env[KIMI_LEGACY_ENV] = KIMI_LEGACY_FORCED_VALUE;
+    }
+    // The update preflight runs on the `-p` path too; a plugin spawn must never
+    // swap the certified binary underneath the execution plan.
+    env.KIMI_CODE_NO_AUTO_UPDATE = "1";
+    // Plugin-owned web lifecycles must not inherit ambient idle-session reclamation.
+    env.KIMI_CODE_SESSION_IDLE_RECLAIM_MS = "0";
+    // Rescue and pursue share the write label. Only the immutable execution
+    // plan can grant pursue's narrow goal-metadata access; never trust ambient
+    // state inherited from another plugin job.
+    env.KIMI_PLUGIN_CC_OPERATION = opts.executionPlan.operationKind;
+    env.KIMI_CODE_HOME = resolveKimiHome(opts.env, opts.cwd);
+    if (opts.commandLabel !== undefined) {
+        env.KIMI_PLUGIN_CC_CMD = opts.commandLabel;
+    }
+    if (opts.swarmMaxConcurrency !== undefined) {
+        env.KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY = String(opts.swarmMaxConcurrency);
+    }
+    if (opts.trustedWorkspaceRoot !== undefined) {
+        // Trusted allowlist root for write-capable PreToolUse cases. Forge-proof
+        // (env on the spawned process; the model cannot alter it). See field doc.
+        env.KIMI_PLUGIN_CC_WORKSPACE_ROOT = opts.trustedWorkspaceRoot;
+    }
+    return env;
+}
+function appendToTail(tail, chunk, maxBytes) {
+    const combined = tail + chunk;
+    if (combined.length <= maxBytes)
+        return combined;
+    return combined.slice(combined.length - maxBytes);
+}
+function truncateChars(value, maxChars) {
+    return value.length <= maxChars ? value : value.slice(0, maxChars);
+}
+function isErrnoException(err, code) {
+    return typeof err === "object" && err !== null && "code" in err && err.code === code;
+}
+function raceWithTimeout(promise, timeoutMs) {
+    let timer;
+    const timeoutPromise = new Promise((resolve) => {
+        timer = setTimeout(resolve, timeoutMs);
+    });
+    return Promise.race([
+        promise.then(() => { }, () => { }),
+        timeoutPromise,
+    ]).finally(() => {
+        if (timer !== undefined)
+            clearTimeout(timer);
+    });
+}
+async function raceWithValueTimeout(promise, timeoutMs, timeoutValue) {
+    let timer;
+    try {
+        return await Promise.race([
+            promise,
+            new Promise((resolve) => {
+                timer = setTimeout(() => resolve(timeoutValue), Math.max(0, timeoutMs));
+            }),
+        ]);
+    }
+    finally {
+        if (timer !== undefined)
+            clearTimeout(timer);
+    }
+}