chore: update lockfile

Co-Authored-By: Paperclip <noreply@paperclip.ing>

dist/server/execute.d.ts

@@ -1,3 +1,36 @@
 import type { AdapterExecutionContext, AdapterExecutionResult } from "@paperclipai/adapter-utils";
+import type * as k8s from "@kubernetes/client-node";
+/**
+ * Detect a Kubernetes 404 (Not Found) error from @kubernetes/client-node.
+ * Works for both v0.x (response.statusCode) and v1.0+ (response.status, message).
+ * Exported for unit tests.
+ */
+export declare function isK8s404(err: unknown): boolean;
+/**
+ * Build the error message when Claude's stdout contains no result event.
+ * Skips system/init event lines so the UI doesn't display the raw init JSON.
+ * Exported for unit tests.
+ */
+export declare function buildPartialRunError(exitCode: number | null, model: string, stdout: string): string;
+/**
+ * Evaluate an orphaned K8s Job (one whose `paperclip.io/run-id` label does
+ * not match the current runId) as a potential reattach target.  A Job is
+ * reattachable when it belongs to the same agent, same task, and same resume
+ * session as the current run — meaning the previous Paperclip instance was
+ * mid-stream on the exact piece of work this new run was dispatched to do.
+ * Exported for unit tests.
+ */
+export declare function isReattachableOrphan(job: k8s.V1Job, expected: {
+    agentId: string;
+    taskId: string | null;
+    sessionId: string | null;
+}): boolean;
+/**
+ * Build an error message for a pod that reached phase=Failed before or
+ * instead of streaming logs. Includes the claude container's terminated exit
+ * code and reason when available so operators can diagnose crashes without
+ * needing kubectl.  Exported for unit tests.
+ */
+export declare function describePodTerminatedError(podName: string, phase: string, containerStatuses: k8s.V1ContainerStatus[]): string;
 export declare function execute(ctx: AdapterExecutionContext): Promise<AdapterExecutionResult>;
 //# sourceMappingURL=execute.d.ts.map

dist/server/execute.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"execute.d.ts","sourceRoot":"","sources":["../../src/server/execute.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,uBAAuB,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AAiUlG,wBAAsB,OAAO,CAAC,GAAG,EAAE,uBAAuB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAoc3F"}
+{"version":3,"file":"execute.d.ts","sourceRoot":"","sources":["../../src/server/execute.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,uBAAuB,EAAE,sBAAsB,EAAE,MAAM,4BAA4B,CAAC;AAWlG,OAAO,KAAK,KAAK,GAAG,MAAM,yBAAyB,CAAC;AAYpD;;;;GAIG;AACH,wBAAgB,QAAQ,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAO9C;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,MAAM,GAAG,IAAI,EACvB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,GACb,MAAM,CA4BR;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,GAAG,EAAE,GAAG,CAAC,KAAK,EACd,QAAQ,EAAE;IAAE,OAAO,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAAC,SAAS,EAAE,MAAM,GAAG,IAAI,CAAA;CAAE,GAC7E,OAAO,CAaT;AAED;;;;;GAKG;AACH,wBAAgB,0BAA0B,CACxC,OAAO,EAAE,MAAM,EACf,KAAK,EAAE,MAAM,EACb,iBAAiB,EAAE,GAAG,CAAC,iBAAiB,EAAE,GACzC,MAAM,CASR;AAkWD,wBAAsB,OAAO,CAAC,GAAG,EAAE,uBAAuB,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAkkB3F"}

dist/server/execute.js

@@ -1,12 +1,110 @@
 import { asString, asNumber, asBoolean, parseObject } from "@paperclipai/adapter-utils/server-utils";
 import { parseClaudeStreamJson, describeClaudeFailure, isClaudeMaxTurnsResult, isClaudeUnknownSessionError, } from "./parse.js";
 import { getSelfPodInfo, getBatchApi, getCoreApi, getLogApi } from "./k8s-client.js";
-import { buildJobManifest } from "./job-manifest.js";
+import { buildJobManifest, sanitizeLabelValue } from "./job-manifest.js";
+import { LogLineDedupFilter } from "./log-dedup.js";
 import { Writable } from "node:stream";
 const POLL_INTERVAL_MS = 2000;
 const KEEPALIVE_INTERVAL_MS = 15_000;
 const LOG_STREAM_RECONNECT_DELAY_MS = 3_000;
 const MAX_LOG_RECONNECT_ATTEMPTS = 50;
+// How long to keep refreshing onSpawn after the Job reaches a terminal state.
+// Covers the cleanup path (delete job, parse stdout) so a slow K8s API call
+// doesn't trip the 5-minute reaper staleness window.
+const POST_TERMINAL_KEEPALIVE_MS = 90_000;
+/**
+ * Detect a Kubernetes 404 (Not Found) error from @kubernetes/client-node.
+ * Works for both v0.x (response.statusCode) and v1.0+ (response.status, message).
+ * Exported for unit tests.
+ */
+export function isK8s404(err) {
+    if (!(err instanceof Error))
+        return false;
+    const e = err;
+    const resp = e.response;
+    if (resp?.statusCode === 404 || resp?.status === 404)
+        return true;
+    if (e.statusCode === 404)
+        return true;
+    return /HTTP-Code:\s*404\b/.test(err.message);
+}
+/**
+ * Build the error message when Claude's stdout contains no result event.
+ * Skips system/init event lines so the UI doesn't display the raw init JSON.
+ * Exported for unit tests.
+ */
+export function buildPartialRunError(exitCode, model, stdout) {
+    if (exitCode === 0)
+        return "Failed to parse Claude JSON output";
+    // Walk stdout lines, skip system events, return the first real content line.
+    const firstContentLine = stdout.split(/\r?\n/)
+        .map((l) => l.trim())
+        .find((l) => {
+        if (!l)
+            return false;
+        try {
+            const obj = JSON.parse(l);
+            if (typeof obj === "object" && obj !== null && obj.type === "system")
+                return false;
+        }
+        catch {
+            // not JSON — treat as content
+        }
+        return true;
+    }) ?? "";
+    // If we only have system/init events and nothing else, surface the model
+    // name so the operator can diagnose missing credentials or unsupported model.
+    const initOnlyOutput = stdout.trim() !== "" && model !== "" && !firstContentLine;
+    if (initOnlyOutput) {
+        const modelHint = model ? ` (model: ${model})` : "";
+        return `Claude started but did not produce a result${modelHint} — check API credentials, model support, and adapter config`;
+    }
+    return firstContentLine
+        ? `Claude exited with code ${exitCode ?? -1}: ${firstContentLine}`
+        : `Claude exited with code ${exitCode ?? -1}`;
+}
+/**
+ * Evaluate an orphaned K8s Job (one whose `paperclip.io/run-id` label does
+ * not match the current runId) as a potential reattach target.  A Job is
+ * reattachable when it belongs to the same agent, same task, and same resume
+ * session as the current run — meaning the previous Paperclip instance was
+ * mid-stream on the exact piece of work this new run was dispatched to do.
+ * Exported for unit tests.
+ */
+export function isReattachableOrphan(job, expected) {
+    if (!expected.taskId || !expected.sessionId)
+        return false;
+    const labels = job.metadata?.labels ?? {};
+    if (labels["paperclip.io/adapter-type"] !== "claude_k8s")
+        return false;
+    if (labels["paperclip.io/agent-id"] !== expected.agentId)
+        return false;
+    if (labels["paperclip.io/task-id"] !== expected.taskId)
+        return false;
+    if (labels["paperclip.io/session-id"] !== expected.sessionId)
+        return false;
+    const conditions = job.status?.conditions ?? [];
+    const terminal = conditions.some((c) => (c.type === "Complete" || c.type === "Failed") && c.status === "True");
+    if (terminal)
+        return false;
+    return true;
+}
+/**
+ * Build an error message for a pod that reached phase=Failed before or
+ * instead of streaming logs. Includes the claude container's terminated exit
+ * code and reason when available so operators can diagnose crashes without
+ * needing kubectl.  Exported for unit tests.
+ */
+export function describePodTerminatedError(podName, phase, containerStatuses) {
+    const mainCs = containerStatuses.find((cs) => cs.name === "claude");
+    const terminated = mainCs?.state?.terminated;
+    if (terminated) {
+        const code = terminated.exitCode ?? "unknown";
+        const reason = terminated.reason ?? terminated.message ?? "no reason";
+        return `Pod ${podName} reached phase=${phase}: claude exited ${code} (${reason})`;
+    }
+    return `Pod ${podName} reached phase=${phase}`;
+}
 /**
  * Wait for the Job's pod to reach a terminal or running state.
  * Returns the pod name once logs can be streamed, or throws on failure.
@@ -52,14 +150,22 @@ async function waitForPod(namespace, jobName, timeoutMs, onLog, kubeconfigPath)
                     details.push(`${cs.name}: waiting (${cs.state.waiting.reason ?? "unknown"})`);
                 else if (cs.state?.running)
                     details.push(`${cs.name}: running`);
+                else if (cs.state?.terminated)
+                    details.push(`${cs.name}: terminated (exit ${cs.state.terminated.exitCode ?? "?"}, ${cs.state.terminated.reason ?? "no reason"})`);
             }
             await onLog("stdout", `[paperclip] Pod ${podName}: ${details.join(", ")}\n`);
             lastStatus = statusKey;
         }
         // Ready to stream logs
-        if (phase === "Running" || phase === "Succeeded" || phase === "Failed") {
+        if (phase === "Running" || phase === "Succeeded") {
             return podName;
         }
+        // phase=Failed means the pod crashed before we could stream logs.
+        // Throwing here routes the caller into the error path with a structured
+        // message instead of entering the log-streaming path with a dead pod.
+        if (phase === "Failed") {
+            throw new Error(describePodTerminatedError(podName, phase, containerStatuses));
+        }
         // Init containers done + main running (phase may still say Pending briefly)
         const allInitsDone = initStatuses.length > 0 && initStatuses.every((s) => s.state?.terminated?.exitCode === 0);
         const mainRunning = containerStatuses.some((s) => s.state?.running);
@@ -104,16 +210,32 @@ async function waitForPod(namespace, jobName, timeoutMs, onLog, kubeconfigPath)
  * Stream pod logs once via follow. Returns accumulated stdout when the
  * stream ends (container exit, API disconnect, or abort signal).
  */
-async function streamPodLogsOnce(namespace, podName, onLog, kubeconfigPath, sinceSeconds) {
+async function streamPodLogsOnce(namespace, podName, onLog, kubeconfigPath, sinceSeconds, dedup, stopSignal) {
     const logApi = getLogApi(kubeconfigPath);
     const chunks = [];
     const writable = new Writable({
         write(chunk, _encoding, callback) {
             const text = chunk.toString("utf-8");
             chunks.push(text);
-            void onLog("stdout", text).then(() => callback(), callback);
+            const emitted = dedup ? dedup.filter(text) : text;
+            if (!emitted) {
+                callback();
+                return;
+            }
+            void onLog("stdout", emitted).then(() => callback(), callback);
         },
     });
+    // When the job completion signal fires, destroy the writable to abort the
+    // in-flight follow stream.  Without this, logApi.log can hang indefinitely
+    // when the pod terminates without closing the HTTP connection cleanly.
+    let stopPoller = null;
+    if (stopSignal) {
+        stopPoller = setInterval(() => {
+            if (stopSignal.stopped && !writable.destroyed) {
+                writable.destroy();
+            }
+        }, 200);
+    }
     try {
         await logApi.log(namespace, podName, "claude", writable, {
             follow: true,
@@ -122,8 +244,12 @@ async function streamPodLogsOnce(namespace, podName, onLog, kubeconfigPath, sinc
         });
     }
     catch {
-        // follow may fail if the container already exited or the API
-        // connection dropped — not fatal, caller decides whether to retry.
+        // follow may fail if the container already exited, the API connection
+        // dropped, or we aborted via writable.destroy() — not fatal.
+    }
+    finally {
+        if (stopPoller)
+            clearInterval(stopPoller);
     }
     return chunks.join("");
 }
@@ -143,6 +269,9 @@ async function streamPodLogs(namespace, podName, onLog, kubeconfigPath, stopSign
     // reconnects use a tight window instead of an ever-growing one anchored
     // at stream start.  This is the primary fix for FAR-105 duplicative logs.
     let lastLogReceivedAt = Math.floor(Date.now() / 1000);
+    // Shared across reconnects so replayed lines inside the `sinceSeconds`
+    // overlap window are dropped before they reach the streaming UI (FAR-123).
+    const dedup = new LogLineDedupFilter();
     while (!stopSignal?.stopped) {
         if (attempt >= MAX_LOG_RECONNECT_ATTEMPTS) {
             await onLog("stderr", `[paperclip] Log stream: max reconnect attempts (${MAX_LOG_RECONNECT_ATTEMPTS}) reached — giving up.\n`);
@@ -158,7 +287,7 @@ async function streamPodLogs(namespace, podName, onLog, kubeconfigPath, stopSign
             await onLog("stdout", `[paperclip] Log stream disconnected — reconnecting (attempt ${attempt}/${MAX_LOG_RECONNECT_ATTEMPTS})...\n`);
         }
         const preStreamTs = Math.floor(Date.now() / 1000);
-        const result = await streamPodLogsOnce(namespace, podName, onLog, kubeconfigPath, sinceSeconds);
+        const result = await streamPodLogsOnce(namespace, podName, onLog, kubeconfigPath, sinceSeconds, dedup, stopSignal);
         if (result) {
             allChunks.push(result);
             // Update last-received timestamp to now (the stream just ended,
@@ -177,6 +306,11 @@ async function streamPodLogs(namespace, podName, onLog, kubeconfigPath, stopSign
         // Brief pause before reconnecting to avoid tight loops.
         await new Promise((resolve) => setTimeout(resolve, LOG_STREAM_RECONNECT_DELAY_MS));
     }
+    // Flush any buffered partial line so the final assistant/result chunk
+    // isn't dropped when the stream ends mid-line.
+    const tail = dedup.flush();
+    if (tail)
+        await onLog("stdout", tail);
     return allChunks.join("");
 }
 /**
@@ -199,13 +333,27 @@ async function readPodLogs(namespace, podName, kubeconfigPath) {
 }
 /**
  * Wait for the Job to reach a terminal state (Complete or Failed).
- * Returns the Job's final status.
+ * Returns the Job's final status.  A 404 (job deleted by TTL or externally)
+ * is treated as a soft terminal: succeeded=false, timedOut=false, jobGone=true.
+ * The caller should log this and fall through to stdout parsing.
  */
 async function waitForJobCompletion(namespace, jobName, timeoutMs, kubeconfigPath) {
     const batchApi = getBatchApi(kubeconfigPath);
     const deadline = timeoutMs > 0 ? Date.now() + timeoutMs : 0;
     while (deadline === 0 || Date.now() < deadline) {
-        const job = await batchApi.readNamespacedJob({ name: jobName, namespace });
+        let job;
+        try {
+            job = await batchApi.readNamespacedJob({ name: jobName, namespace });
+        }
+        catch (err) {
+            if (isK8s404(err)) {
+                // Job was deleted (TTL garbage collection or external deletion) before
+                // we detected its terminal condition.  The container must have already
+                // exited for TTL to fire, so log streaming will have captured the output.
+                return { succeeded: false, timedOut: false, jobGone: true };
+            }
+            throw err;
+        }
         const conditions = job.status?.conditions ?? [];
         const complete = conditions.find((c) => c.type === "Complete" && c.status === "True");
         if (complete)
@@ -261,10 +409,18 @@ export async function execute(ctx) {
     // Guard: claude_k8s must not run concurrently for the same agent (shared PVC/session).
     // After a server restart, orphaned K8s Jobs from previous (now-failed) runs may
     // still be running.  We detect those by comparing the Job's run-id label against
-    // the current runId and clean them up so this execution can proceed.
+    // the current runId.  When reattachOrphanedJobs is enabled and the orphan matches
+    // the current agent+task+session, we attach to it instead of deleting it (FAR-124).
     const agentId = ctx.agent.id;
     const selfPod = await getSelfPodInfo(kubeconfigPath);
     const guardNamespace = asString(config.namespace, "") || selfPod.namespace;
+    const reattachOrphanedJobs = asBoolean(config.reattachOrphanedJobs, true);
+    const runtimeSessionParams = parseObject(runtime.sessionParams);
+    const currentSessionIdRaw = asString(runtimeSessionParams.sessionId, runtime.sessionId ?? "");
+    const currentSessionLabel = currentSessionIdRaw ? sanitizeLabelValue(currentSessionIdRaw) : null;
+    const currentTaskIdRaw = asString(ctx.context.taskId, "") || asString(ctx.context.issueId, "");
+    const currentTaskLabel = currentTaskIdRaw ? sanitizeLabelValue(currentTaskIdRaw) : null;
+    let reattachTarget = null;
     try {
         const batchApi = getBatchApi(kubeconfigPath);
         const existing = await batchApi.listNamespacedJob({
@@ -277,10 +433,37 @@ export async function execute(ctx) {
             // concurrent jobs (same runId — shouldn't happen but guard defensively).
             const orphaned = running.filter((j) => (j.metadata?.labels?.["paperclip.io/run-id"] ?? "") !== runId);
             const samRun = running.filter((j) => (j.metadata?.labels?.["paperclip.io/run-id"] ?? "") === runId);
-            if (orphaned.length > 0) {
-                const orphanNames = orphaned.map((j) => j.metadata?.name).join(", ");
-                await onLog("stdout", `[paperclip] Cleaning up ${orphaned.length} orphaned K8s Job(s) from previous run(s): ${orphanNames}\n`);
-                for (const j of orphaned) {
+            // Pick the most recent reattachable orphan — same agent + task + session,
+            // not terminal.  Only one target is chosen; any other orphans get
+            // cleaned up as before.
+            if (reattachOrphanedJobs && orphaned.length > 0) {
+                const candidates = orphaned
+                    .filter((j) => isReattachableOrphan(j, {
+                    agentId,
+                    taskId: currentTaskLabel,
+                    sessionId: currentSessionLabel,
+                }))
+                    .sort((a, b) => {
+                    const at = new Date(a.metadata?.creationTimestamp ?? 0).getTime();
+                    const bt = new Date(b.metadata?.creationTimestamp ?? 0).getTime();
+                    return bt - at;
+                });
+                const chosen = candidates[0];
+                const chosenName = chosen?.metadata?.name;
+                if (chosen && chosenName) {
+                    reattachTarget = {
+                        jobName: chosenName,
+                        namespace: chosen.metadata?.namespace ?? guardNamespace,
+                        priorRunId: chosen.metadata?.labels?.["paperclip.io/run-id"] ?? "",
+                        image: chosen.spec?.template?.spec?.containers?.[0]?.image ?? "unknown",
+                    };
+                }
+            }
+            const toDelete = orphaned.filter((j) => !reattachTarget || j.metadata?.name !== reattachTarget.jobName);
+            if (toDelete.length > 0) {
+                const orphanNames = toDelete.map((j) => j.metadata?.name).join(", ");
+                await onLog("stdout", `[paperclip] Cleaning up ${toDelete.length} orphaned K8s Job(s) from previous run(s): ${orphanNames}\n`);
+                for (const j of toDelete) {
                     const name = j.metadata?.name;
                     if (name) {
                         await cleanupJob(guardNamespace, name, onLog, kubeconfigPath);
@@ -317,81 +500,114 @@ export async function execute(ctx) {
             errorCode: "k8s_concurrency_guard_unreachable",
         };
     }
-    // Build Job manifest
-    const { job, jobName, namespace, prompt, claudeArgs, promptMetrics, promptSecret } = buildJobManifest({
-        ctx,
-        selfPod,
-    });
-    // Report invocation metadata
-    if (onMeta) {
-        await onMeta({
-            adapterType: "claude_k8s",
-            command: `kubectl job/${jobName}`,
-            cwd: namespace,
-            commandArgs: claudeArgs,
-            commandNotes: [
-                `Image: ${job.spec?.template.spec?.containers[0]?.image ?? "unknown"}`,
-                `Namespace: ${namespace}`,
-                `Timeout: ${timeoutSec}s`,
-            ],
-            prompt,
-            ...(promptMetrics ? { promptMetrics } : {}),
-            context: ctx.context,
-        });
-    }
-    // If the prompt is large, create a Secret to hold it (avoids the ~1 MiB
-    // PodSpec limit).  The Secret is cleaned up in the finally block.
     const coreApi = getCoreApi(kubeconfigPath);
-    if (promptSecret) {
-        try {
-            await coreApi.createNamespacedSecret({
-                namespace: promptSecret.namespace,
-                body: {
-                    apiVersion: "v1",
-                    kind: "Secret",
-                    metadata: {
-                        name: promptSecret.name,
-                        namespace: promptSecret.namespace,
-                        labels: {
-                            "app.kubernetes.io/managed-by": "paperclip",
-                            "paperclip.io/adapter-type": "claude_k8s",
-                            "paperclip.io/run-id": runId,
-                        },
-                    },
-                    stringData: promptSecret.data,
-                },
+    const batchApi = getBatchApi(kubeconfigPath);
+    let jobName;
+    let namespace;
+    let promptSecret = null;
+    if (reattachTarget) {
+        jobName = reattachTarget.jobName;
+        namespace = reattachTarget.namespace;
+        // Announce reattach metadata.  Prompt and args aren't known here — they
+        // belong to the prior run that created this pod and are already present
+        // on the running container.
+        if (onMeta) {
+            await onMeta({
+                adapterType: "claude_k8s",
+                command: `kubectl job/${jobName}`,
+                cwd: namespace,
+                commandArgs: [],
+                commandNotes: [
+                    `Image: ${reattachTarget.image}`,
+                    `Namespace: ${namespace}`,
+                    `Reattached from prior run: ${reattachTarget.priorRunId || "unknown"}`,
+                    `Timeout: ${timeoutSec}s`,
+                ],
+                prompt: "",
+                context: ctx.context,
             });
-            await onLog("stdout", `[paperclip] Created prompt Secret: ${promptSecret.name} (${Math.round(Buffer.byteLength(prompt, "utf-8") / 1024)} KiB)\n`);
+        }
+        await onLog("stdout", `[paperclip] Reattaching to in-flight K8s Job ${jobName} in namespace ${namespace} (prior run ${reattachTarget.priorRunId || "unknown"})\n`);
+    }
+    else {
+        // Build Job manifest
+        const built = buildJobManifest({ ctx, selfPod });
+        const job = built.job;
+        jobName = built.jobName;
+        namespace = built.namespace;
+        const prompt = built.prompt;
+        const claudeArgs = built.claudeArgs;
+        const promptMetrics = built.promptMetrics;
+        promptSecret = built.promptSecret;
+        // Report invocation metadata
+        if (onMeta) {
+            await onMeta({
+                adapterType: "claude_k8s",
+                command: `kubectl job/${jobName}`,
+                cwd: namespace,
+                commandArgs: claudeArgs,
+                commandNotes: [
+                    `Image: ${job.spec?.template.spec?.containers[0]?.image ?? "unknown"}`,
+                    `Namespace: ${namespace}`,
+                    `Timeout: ${timeoutSec}s`,
+                ],
+                prompt,
+                ...(promptMetrics ? { promptMetrics } : {}),
+                context: ctx.context,
+            });
+        }
+        // If the prompt is large, create a Secret to hold it (avoids the ~1 MiB
+        // PodSpec limit).  The Secret is cleaned up in the finally block.
+        if (promptSecret) {
+            try {
+                await coreApi.createNamespacedSecret({
+                    namespace: promptSecret.namespace,
+                    body: {
+                        apiVersion: "v1",
+                        kind: "Secret",
+                        metadata: {
+                            name: promptSecret.name,
+                            namespace: promptSecret.namespace,
+                            labels: {
+                                "app.kubernetes.io/managed-by": "paperclip",
+                                "paperclip.io/adapter-type": "claude_k8s",
+                                "paperclip.io/run-id": runId,
+                            },
+                        },
+                        stringData: promptSecret.data,
+                    },
+                });
+                await onLog("stdout", `[paperclip] Created prompt Secret: ${promptSecret.name} (${Math.round(Buffer.byteLength(prompt, "utf-8") / 1024)} KiB)\n`);
+            }
+            catch (err) {
+                const msg = err instanceof Error ? err.message : String(err);
+                await onLog("stderr", `[paperclip] Failed to create prompt Secret: ${msg}\n`);
+                return {
+                    exitCode: null,
+                    signal: null,
+                    timedOut: false,
+                    errorMessage: `Failed to create prompt Secret: ${msg}`,
+                    errorCode: "k8s_prompt_secret_create_failed",
+                };
+            }
+        }
+        // Create the Job
+        try {
+            await batchApi.createNamespacedJob({ namespace, body: job });
         }
         catch (err) {
             const msg = err instanceof Error ? err.message : String(err);
-            await onLog("stderr", `[paperclip] Failed to create prompt Secret: ${msg}\n`);
+            await onLog("stderr", `[paperclip] Failed to create K8s Job: ${msg}\n`);
             return {
                 exitCode: null,
                 signal: null,
                 timedOut: false,
-                errorMessage: `Failed to create prompt Secret: ${msg}`,
-                errorCode: "k8s_prompt_secret_create_failed",
+                errorMessage: `Failed to create Kubernetes Job: ${msg}`,
+                errorCode: "k8s_job_create_failed",
             };
         }
+        await onLog("stdout", `[paperclip] Created K8s Job: ${jobName} in namespace ${namespace} (deadline: ${timeoutSec > 0 ? `${timeoutSec}s` : "none"})\n`);
     }
-    // Create the Job
-    const batchApi = getBatchApi(kubeconfigPath);
-    try {
-        await batchApi.createNamespacedJob({ namespace, body: job });
-    }
-    catch (err) {
-        const msg = err instanceof Error ? err.message : String(err);
-        await onLog("stderr", `[paperclip] Failed to create K8s Job: ${msg}\n`);
-        return {
-            exitCode: null,
-            signal: null,
-            timedOut: false,
-            errorMessage: `Failed to create Kubernetes Job: ${msg}`,
-            errorCode: "k8s_job_create_failed",
-        };
-    }
-    await onLog("stdout", `[paperclip] Created K8s Job: ${jobName} in namespace ${namespace} (deadline: ${timeoutSec > 0 ? `${timeoutSec}s` : "none"})\n`);
     let stdout = "";
     let exitCode = null;
     let jobTimedOut = false;
@@ -404,8 +620,24 @@ export async function execute(ctx) {
         const scheduleTimeoutMs = 120_000; // 2 minutes for scheduling
         let podName;
         try {
-            podName = await waitForPod(namespace, jobName, scheduleTimeoutMs, onLog, kubeconfigPath);
-            await onLog("stdout", `[paperclip] Pod running: ${podName}\n`);
+            if (reattachTarget) {
+                // Pod is already running from the prior run — look it up directly.
+                const podList = await coreApi.listNamespacedPod({
+                    namespace,
+                    labelSelector: `job-name=${jobName}`,
+                });
+                const pod = podList.items[0];
+                const name = pod?.metadata?.name;
+                if (!name) {
+                    throw new Error(`Reattach target Job ${jobName} has no pod`);
+                }
+                podName = name;
+                await onLog("stdout", `[paperclip] Reattached to pod ${podName}\n`);
+            }
+            else {
+                podName = await waitForPod(namespace, jobName, scheduleTimeoutMs, onLog, kubeconfigPath);
+                await onLog("stdout", `[paperclip] Pod running: ${podName}\n`);
+            }
             // Notify the server that execution has started.  This sets
             // processStartedAt and refreshes updatedAt in the DB, which the
             // stale-run reaper (reapOrphanedRuns) uses to decide liveness.
@@ -419,13 +651,14 @@ export async function execute(ctx) {
         }
         catch (err) {
             const msg = err instanceof Error ? err.message : String(err);
-            await onLog("stderr", `[paperclip] Pod scheduling failed: ${msg}\n`);
+            const phase = reattachTarget ? "reattach" : "scheduling";
+            await onLog("stderr", `[paperclip] Pod ${phase} failed: ${msg}\n`);
             return {
                 exitCode: null,
                 signal: null,
                 timedOut: false,
-                errorMessage: `Pod scheduling failed: ${msg}`,
-                errorCode: "k8s_pod_schedule_failed",
+                errorMessage: `Pod ${phase} failed: ${msg}`,
+                errorCode: reattachTarget ? "k8s_pod_reattach_failed" : "k8s_pod_schedule_failed",
             };
         }
         // Stream logs and wait for completion concurrently.
@@ -457,18 +690,32 @@ export async function execute(ctx) {
         let lastLogAt = Date.now();
         let keepaliveTick = 0;
         let keepaliveJobTerminal = false;
+        let keepaliveJobTerminalAt = null;
         keepaliveTimer = setInterval(() => {
             // Fire-and-forget the async work; setInterval callbacks must be
             // synchronous or the timer will drift.
             void (async () => {
-                if (keepaliveJobTerminal)
+                if (keepaliveJobTerminal) {
+                    // Post-terminal window: keep refreshing onSpawn during cleanup
+                    // (job deletion, log parsing, K8s API calls) so the reaper doesn't
+                    // fire a false process_lost while execute() is still running.
+                    if (ctx.onSpawn &&
+                        keepaliveJobTerminalAt !== null &&
+                        Date.now() - keepaliveJobTerminalAt <= POST_TERMINAL_KEEPALIVE_MS) {
+                        keepaliveTick++;
+                        if (keepaliveTick % 6 === 0) {
+                            void ctx.onSpawn({ pid: process.pid, processGroupId: null, startedAt: new Date().toISOString() }).catch(() => { });
+                        }
+                    }
                     return;
+                }
                 // Verify the Job is still alive before announcing or refreshing.
                 try {
                     const job = await batchApi.readNamespacedJob({ name: jobName, namespace });
                     const terminal = job.status?.conditions?.some((c) => (c.type === "Complete" || c.type === "Failed") && c.status === "True");
                     if (terminal) {
                         keepaliveJobTerminal = true;
+                        keepaliveJobTerminalAt = Date.now();
                         return;
                     }
                 }
@@ -477,10 +724,9 @@ export async function execute(ctx) {
                     // connection resets should NOT permanently disable the keepalive —
                     // the next tick will re-check and the reaper uses the staleness
                     // window as a safety net.
-                    const statusCode = err?.response?.statusCode
-                        ?? err?.statusCode;
-                    if (statusCode === 404) {
+                    if (isK8s404(err)) {
                         keepaliveJobTerminal = true;
+                        keepaliveJobTerminalAt = Date.now();
                         return;
                     }
                     // Log transient errors but leave keepaliveJobTerminal false so
@@ -525,23 +771,44 @@ export async function execute(ctx) {
         if (logResult.status === "fulfilled") {
             stdout = logResult.value;
         }
-        // If the follow stream missed output (container exited quickly), do a
-        // one-shot log read as fallback before the pod is cleaned up.
-        if (!stdout.trim()) {
-            await onLog("stdout", `[paperclip] Log stream returned empty — reading pod logs directly...\n`);
-            stdout = await readPodLogs(namespace, podName, kubeconfigPath);
-            if (stdout.trim()) {
+        // One-shot log fallback: handles two failure modes with a single read.
+        // Mode 1 — empty stream: the follow stream returned nothing (fast exit before connection).
+        // Mode 2 — partial stream: we have some output but no result event (follow stream raced
+        //   with container exit and captured only the init line before the connection dropped).
+        // A one-shot readPodLogs is more reliable for already-terminated containers and reads
+        // from the beginning of the log, giving us the full output.
+        // We use a cheap string scan for the result-event guard (avoids a full JSON parse here;
+        // the authoritative parse happens once below after all fallbacks complete).
+        const hasResultEvent = stdout.includes('"type":"result"');
+        const needsOneShot = !stdout.trim() || (stdout.trim() && !hasResultEvent);
+        if (needsOneShot) {
+            if (!stdout.trim()) {
+                await onLog("stdout", `[paperclip] Log stream returned empty — reading pod logs directly...\n`);
+            }
+            const oneShotLogs = await readPodLogs(namespace, podName, kubeconfigPath);
+            if (!stdout.trim() && oneShotLogs.trim()) {
+                stdout = oneShotLogs;
                 await onLog("stdout", stdout);
             }
+            else if (oneShotLogs && oneShotLogs.length > stdout.length) {
+                await onLog("stdout", `[paperclip] Log stream captured partial output — supplemental one-shot read returned more content.\n`);
+                stdout = oneShotLogs;
+            }
         }
         if (completionResult.status === "fulfilled") {
             jobTimedOut = completionResult.value.timedOut;
+            if (completionResult.value.jobGone) {
+                // Job was deleted by TTL or externally before we observed the Complete/Failed
+                // condition.  The container must have exited first (TTL only fires after
+                // completion), so log streaming has captured the full output — continue
+                // to stdout parsing rather than returning an error.
+                await onLog("stdout", `[paperclip] Job ${jobName} was deleted before terminal condition was observed (TTL or external deletion) — proceeding with captured output.\n`);
+            }
         }
         else {
-            // waitForJobCompletion threw — re-check job state to avoid returning
-            // while the job is still running (which would cause UI staleness and
-            // concurrency errors on retry).  Use a bounded timeout (60s) so we
-            // don't hang the heartbeat indefinitely if the K8s API is degraded.
+            // waitForJobCompletion threw an unexpected error — re-check job state to
+            // avoid returning while the job is still running.  Use a bounded timeout
+            // (60s) so we don't hang the heartbeat indefinitely if the K8s API is degraded.
             jobTimedOut = false;
             const RECHECK_TIMEOUT_MS = 60_000;
             const actualState = await waitForJobCompletion(namespace, jobName, RECHECK_TIMEOUT_MS, kubeconfigPath);
@@ -550,6 +817,11 @@ export async function execute(ctx) {
                 // Return an error so the UI knows the run is not done.
                 jobTimedOut = true;
             }
+            else if (actualState.jobGone) {
+                // Job was deleted before we could confirm terminal state — same as the
+                // fulfilled+jobGone case above: proceed with captured output.
+                await onLog("stdout", `[paperclip] Job ${jobName} was deleted before terminal condition was observed (TTL or external deletion) — proceeding with captured output.\n`);
+            }
             else if (!actualState.succeeded) {
                 // Job still not terminal — the completion error was likely transient.
                 // Return an error so the UI knows the run is not done, rather than
@@ -615,16 +887,11 @@ export async function execute(ctx) {
         };
     }
     if (!parsed) {
-        const stderrLine = stdout.split(/\r?\n/).map((l) => l.trim()).find(Boolean) ?? "";
         return {
             exitCode,
             signal: null,
             timedOut: false,
-            errorMessage: exitCode === 0
-                ? "Failed to parse Claude JSON output"
-                : stderrLine
-                    ? `Claude exited with code ${exitCode ?? -1}: ${stderrLine}`
-                    : `Claude exited with code ${exitCode ?? -1}`,
+            errorMessage: buildPartialRunError(exitCode, parsedStream.model, stdout),
             resultJson: { stdout },
         };
     }
@@ -636,8 +903,7 @@ export async function execute(ctx) {
             outputTokens: asNumber(usageObj.output_tokens, 0),
         };
     })();
-    const runtimeSessionParams = parseObject(runtime.sessionParams);
-    const fallbackSessionId = asString(runtimeSessionParams.sessionId, runtime.sessionId ?? "");
+    const fallbackSessionId = currentSessionIdRaw;
     const resolvedSessionId = parsedStream.sessionId
         ?? (asString(parsed.session_id, fallbackSessionId) || fallbackSessionId);
     const model = asString(config.model, "");

dist/server/job-manifest.d.ts

@@ -1,5 +1,19 @@
 import type * as k8s from "@kubernetes/client-node";
 import type { AdapterExecutionContext } from "@paperclipai/adapter-utils";
+/**
+ * Build the shell command prefix that installs a native Node.js PostToolUse
+ * hook into Claude Code's settings.  The hook truncates oversized tool outputs
+ * before they reach the model — replacing the RTK binary init-container
+ * approach with a self-contained Node.js implementation.
+ *
+ * Both scripts are base64-encoded so they can be embedded in a sh -c command
+ * string without any quoting or escaping issues.
+ *
+ * @param maxOutputBytes  Byte threshold above which tool output is truncated.
+ * @returns               A shell command string (suitable for "&&"-chaining
+ *                        before the claude invocation).
+ */
+export declare function buildRtkSetupCommands(maxOutputBytes: number): string;
 import type { SelfPodInfo } from "./k8s-client.js";
 export interface JobBuildInput {
     ctx: AdapterExecutionContext;
@@ -24,5 +38,11 @@ export interface JobBuildResult {
      *  staged as a K8s Secret before creating the Job. */
     promptSecret: PromptSecret | null;
 }
+/**
+ * Sanitize a string for use as a Kubernetes label value (RFC 1123 subset:
+ * `[a-zA-Z0-9]([-_.a-zA-Z0-9]*[a-zA-Z0-9])?`, max 63 chars).  Returns `null`
+ * when no usable characters remain — the caller should omit the label.
+ */
+export declare function sanitizeLabelValue(value: string, maxLen?: number): string | null;
 export declare function buildJobManifest(input: JobBuildInput): JobBuildResult;
 //# sourceMappingURL=job-manifest.d.ts.map

dist/server/job-manifest.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"job-manifest.d.ts","sourceRoot":"","sources":["../../src/server/job-manifest.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,KAAK,GAAG,MAAM,yBAAyB,CAAC;AACpD,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,4BAA4B,CAAC;AAgD1E,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AA6CnD,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,uBAAuB,CAAC;IAC7B,OAAO,EAAE,WAAW,CAAC;CACtB;AAED;;+EAE+E;AAC/E,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC9B;AAED,MAAM,WAAW,cAAc;IAC7B,GAAG,EAAE,GAAG,CAAC,KAAK,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACtC;0DACsD;IACtD,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;CACnC;AAuHD,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,aAAa,GAAG,cAAc,CAkRrE"}
+{"version":3,"file":"job-manifest.d.ts","sourceRoot":"","sources":["../../src/server/job-manifest.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,KAAK,GAAG,MAAM,yBAAyB,CAAC;AACpD,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,4BAA4B,CAAC;AAY1E;;;;;;;;;;;;GAYG;AACH,wBAAgB,qBAAqB,CAAC,cAAc,EAAE,MAAM,GAAG,MAAM,CAiEpE;AAsCD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AA6CnD,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,uBAAuB,CAAC;IAC7B,OAAO,EAAE,WAAW,CAAC;CACtB;AAED;;+EAE+E;AAC/E,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC9B;AAED,MAAM,WAAW,cAAc;IAC7B,GAAG,EAAE,GAAG,CAAC,KAAK,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACtC;0DACsD;IACtD,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;CACnC;AAMD;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,SAAK,GAAG,MAAM,GAAG,IAAI,CAI5E;AAmHD,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,aAAa,GAAG,cAAc,CAmSrE"}

dist/server/job-manifest.js

@@ -1,5 +1,81 @@
 import { asString, asNumber, asBoolean, asStringArray, parseObject, buildPaperclipEnv, renderTemplate, } from "@paperclipai/adapter-utils/server-utils";
 import { createHash } from "node:crypto";
+/**
+ * Build the shell command prefix that installs a native Node.js PostToolUse
+ * hook into Claude Code's settings.  The hook truncates oversized tool outputs
+ * before they reach the model — replacing the RTK binary init-container
+ * approach with a self-contained Node.js implementation.
+ *
+ * Both scripts are base64-encoded so they can be embedded in a sh -c command
+ * string without any quoting or escaping issues.
+ *
+ * @param maxOutputBytes  Byte threshold above which tool output is truncated.
+ * @returns               A shell command string (suitable for "&&"-chaining
+ *                        before the claude invocation).
+ */
+export function buildRtkSetupCommands(maxOutputBytes) {
+    // --- Filter script ----------------------------------------------------------
+    // This script runs as the PostToolUse hook inside every K8s Job pod.
+    // Claude Code writes the hook event as JSON to the script's stdin; the script
+    // truncates the tool_response/tool_result content when it exceeds the
+    // threshold and writes the (possibly modified) JSON to stdout.
+    //
+    // Field-name coverage:
+    //   • tool_response — documented hook event format for PostToolUse
+    //   • tool_result   — alternative name seen in some Claude Code versions
+    // Content may be a plain string or an array of typed blocks (text/image/…).
+    const filterScript = [
+        `const c=[];`,
+        `process.stdin.on('data',d=>c.push(d));`,
+        `process.stdin.on('end',()=>{`,
+        `const raw=Buffer.concat(c).toString('utf-8');`,
+        `let o;try{o=JSON.parse(raw);}catch{process.stdout.write(raw);return;}`,
+        `const MAX=${maxOutputBytes};`,
+        `function trunc(s){`,
+        `if(typeof s!=='string')return s;`,
+        `const b=Buffer.from(s,'utf-8');`,
+        `if(b.length<=MAX)return s;`,
+        `return b.slice(0,MAX).toString('utf-8')+'\\n[...'+(b.length-MAX)+' bytes truncated by paperclip-rtk]';`,
+        `}`,
+        `const tr=o&&(o.tool_response||o.tool_result);`,
+        `if(tr){`,
+        `if(typeof tr.content==='string'){tr.content=trunc(tr.content);}`,
+        `else if(Array.isArray(tr.content)){`,
+        `tr.content=tr.content.map(function(b){`,
+        `if(b&&typeof b==='object'&&typeof b.text==='string'){`,
+        `return Object.assign({},b,{text:trunc(b.text)});`,
+        `}return b;`,
+        `});`,
+        `}`,
+        `}`,
+        `process.stdout.write(JSON.stringify(o));`,
+        `});`,
+    ].join("");
+    // --- Settings script --------------------------------------------------------
+    // Reads the existing ~/.claude/settings.json (if any), merges in the RTK
+    // PostToolUse hook, and writes the file back.  All other settings sections
+    // are preserved; only PostToolUse is replaced so we own the full hook list
+    // for this run.
+    const settingsScript = [
+        `const fs=require('fs'),pt=require('path');`,
+        `const p=pt.join(process.env.HOME,'.claude','settings.json');`,
+        `let s={};try{s=JSON.parse(fs.readFileSync(p,'utf-8'));}catch(e){}`,
+        `s.hooks=s.hooks||{};`,
+        `s.hooks.PostToolUse=[{matcher:'.*',hooks:[{type:'command',command:'node /tmp/.rtk-filter.js'}]}];`,
+        `fs.mkdirSync(pt.dirname(p),{recursive:true});`,
+        `fs.writeFileSync(p,JSON.stringify(s));`,
+    ].join("");
+    // Encode as base64 so the strings can be embedded directly in a shell command
+    // without any quoting concerns (base64 alphabet: A-Za-z0-9+/=).
+    const filterB64 = Buffer.from(filterScript, "utf-8").toString("base64");
+    const settingsB64 = Buffer.from(settingsScript, "utf-8").toString("base64");
+    return [
+        // Write the filter script
+        `node -e "require('fs').writeFileSync('/tmp/.rtk-filter.js',Buffer.from('${filterB64}','base64').toString('utf-8'))"`,
+        // Install the Claude Code PostToolUse hook (merge into existing settings)
+        `node -e "eval(Buffer.from('${settingsB64}','base64').toString('utf-8'))"`,
+    ].join(" && ");
+}
 /** Prompts above this size (bytes) are staged via a Secret instead of an
  *  init container env var, protecting against the ~1 MiB PodSpec limit. */
 const LARGE_PROMPT_THRESHOLD_BYTES = 256 * 1024;
@@ -91,6 +167,16 @@ function parseKeyValueConfig(raw) {
 function sanitizeForK8sName(value, maxLen = 16) {
     return value.toLowerCase().replace(/[^a-z0-9-]/g, "").slice(0, maxLen);
 }
+/**
+ * Sanitize a string for use as a Kubernetes label value (RFC 1123 subset:
+ * `[a-zA-Z0-9]([-_.a-zA-Z0-9]*[a-zA-Z0-9])?`, max 63 chars).  Returns `null`
+ * when no usable characters remain — the caller should omit the label.
+ */
+export function sanitizeLabelValue(value, maxLen = 63) {
+    const cleaned = value.replace(/[^a-zA-Z0-9._-]/g, "").slice(0, maxLen);
+    const trimmed = cleaned.replace(/^[^a-zA-Z0-9]+/, "").replace(/[^a-zA-Z0-9]+$/, "");
+    return trimmed.length > 0 ? trimmed : null;
+}
 /**
  * Build a short deterministic hash suffix from the raw inputs to avoid
  * collisions when sanitized slugs happen to be identical.
@@ -202,6 +288,8 @@ export function buildJobManifest(input) {
     const nodeSelector = parseKeyValueConfig(config.nodeSelector);
     const tolerations = Array.isArray(config.tolerations) ? config.tolerations : [];
     const extraLabels = parseKeyValueConfig(config.labels);
+    const enableRtk = asBoolean(config.enableRtk, false);
+    const rtkMaxOutputBytes = asNumber(config.rtkMaxOutputBytes, 50000);
     // Resolve working directory — use workspace cwd, fall back to /paperclip
     const workspaceContext = parseObject(context.paperclipWorkspace);
     const workspaceCwd = asString(workspaceContext.cwd, "");
@@ -289,6 +377,17 @@ export function buildJobManifest(input) {
         "paperclip.io/company-id": agent.companyId,
         "paperclip.io/adapter-type": "claude_k8s",
     };
+    // Reattach-target labels: let a future execute() identify this Job as the
+    // continuation of the same logical unit of work (same task + same resume
+    // session) so it can attach to the running pod across a Paperclip restart
+    // instead of deleting it and starting over (FAR-124).
+    const taskIdRaw = asString(context.taskId, "") || asString(context.issueId, "");
+    const taskLabel = taskIdRaw ? sanitizeLabelValue(taskIdRaw) : null;
+    if (taskLabel)
+        labels["paperclip.io/task-id"] = taskLabel;
+    const sessionLabel = runtimeSessionId ? sanitizeLabelValue(runtimeSessionId) : null;
+    if (sessionLabel)
+        labels["paperclip.io/session-id"] = sessionLabel;
     for (const [key, value] of Object.entries(extraLabels)) {
         labels[key] = value;
     }
@@ -345,7 +444,13 @@ export function buildJobManifest(input) {
     };
     // Build the claude command string for the main container
     const claudeArgsEscaped = claudeArgs.map((a) => `'${a.replace(/'/g, "'\\''")}'`).join(" ");
-    const mainCommand = `cat /tmp/prompt/prompt.txt | claude ${claudeArgsEscaped}`;
+    const claudeInvocation = `cat /tmp/prompt/prompt.txt | claude ${claudeArgsEscaped}`;
+    // When RTK output filtering is enabled, prepend the Node.js hook setup.
+    // This writes a filter script and a Claude Code settings file that installs
+    // it as a PostToolUse hook — no external binary or init container required.
+    const mainCommand = enableRtk
+        ? `${buildRtkSetupCommands(rtkMaxOutputBytes)} && ${claudeInvocation}`
+        : claudeInvocation;
     // Decide prompt delivery strategy: env var (small) or Secret volume (large).
     const promptBytes = Buffer.byteLength(prompt, "utf-8");
     const useLargePromptPath = promptBytes > LARGE_PROMPT_THRESHOLD_BYTES;

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "paperclip-adapter-claude-k8s",
-  "version": "0.1.31",
+  "version": "0.1.32",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "paperclip-adapter-claude-k8s",
-      "version": "0.1.31",
+      "version": "0.1.32",
       "license": "MIT",
       "dependencies": {
         "@kubernetes/client-node": "^1.0.0",