@tea-agent/loop-agent 0.39.0-next.26 → 0.39.0-next.28

package/dist/commands/inspect-next.js

@@ -0,0 +1,8 @@
+import { collectNextInfo, formatNextInfo, } from "../../scripts/next-info.mjs";
+export async function runInspectNext(repoRoot, args = []) {
+    const json = args.includes("--json");
+    const result = collectNextInfo({ cwd: repoRoot });
+    process.stdout.write(formatNextInfo(result, { json }));
+    if (!result.ok)
+        process.exitCode = result.exitCode || 1;
+}

package/dist/executors/dag-pi-executor.js

@@ -2,13 +2,14 @@ import path from "node:path";
 import { createHash } from "node:crypto";
 import { readFile } from "node:fs/promises";
 import { writeDagNodeJsonArtifact, writeTextArtifactFile, } from "../infrastructure/harness/artifact-store.js";
-import { executePiStep, } from "./pi-executor.js";
+import { executePiStep, resolvePiBackend, } from "./pi-executor.js";
+import { resolveDagPiExtensions, } from "./pi-extension-resolver.js";
 import { buildPiWriterToolPolicyContext, createPiWriterCustomTools, } from "./pi-writer-tool-policy.js";
 import { cleanupPlaywrightCliDefaultSession, createPlaywrightCliTool, PI_COMMAND_CAPABILITY_REGISTRY, resolveCaseIdFromWriteSet, resolveEvidenceDirFromWriteSet, } from "./pi-playwright-cli-tool.js";
 import { dagCommandPolicyAllows, resolveDagCommandPolicy, } from "../workflows/dag/types.js";
 import { redactPromptForLog, truncateOutput, } from "../shared/output-truncation.js";
 import { GitStatusUnavailableError, pathsChangedDuringRun, readGitStatusPorcelain, recoverRootNulArtifact, snapshotGitStatusPathFingerprints, snapshotGitStatusPorcelain, validateShellWriteGuard, } from "./shell-write-guard.js";
-import { isWriterEmptyDiffRetryCandidate, isWriterTransportRetryCandidate, INCOMPLETE_WRITE_SET_RETRY_CATEGORY, WRITER_CLEAN_TIMEOUT_RETRY_CATEGORY, WRITER_EMPTY_DIFF_RETRY_CATEGORY, } from "../workflows/dag/retry-policy.js";
+import { isTargetTemplateTransientRetryNode, isWriterEmptyDiffRetryCandidate, isWriterTransportRetryCandidate, INCOMPLETE_WRITE_SET_RETRY_CATEGORY, WRITER_CLEAN_TIMEOUT_RETRY_CATEGORY, WRITER_EMPTY_DIFF_RETRY_CATEGORY, } from "../workflows/dag/retry-policy.js";
 import { assessBackendTestMdPlanCompleteness, assessBackendTestMdWriterCompleteness, assessBackendTestPytestPlanCompleteness, assessBackendTestPytestWriterCompleteness, assessBackendTestShardChildCompleteness, backendTestWriterProgressRoleForTask, classifyBackendTestWriterCompletenessFailure, isBackendTestCompletenessRetryCandidate, isBackendTestMdPlanTask, isBackendTestPytestPlanTask, isBackendTestShardChildTask, writeBackendTestWriterProgressArtifacts, } from "../workflows/dag/backend-test-writer-completeness.js";
 import { resolveBackendTestLayout } from "../workflows/dag/backend-test-layout.js";
 import { redactSecrets, truncateUtf8Preview } from "../shared/preview.js";
@@ -327,6 +328,36 @@ export async function writePiExecutorArtifacts(artifactsDir, input) {
     await writeTextArtifactFile(summaryPath, buildPiResultSummaryMarkdown(input));
     return { promptPath, summaryPath };
 }
+/**
+ * Run-owned per-node Pi extension facts (plan 2026-08-21): requested ids,
+ * resolved packages (with absolute entry paths), and degraded/missing ids.
+ * Written only for nodes that declare `piExtensions`; never into the repo-level
+ * DAG template. Best-effort: artifact failures never fail the node.
+ */
+async function writePiExtensionsArtifact(nodeArtifactsDir, requested, resolved, missing) {
+    try {
+        const runDir = path.dirname(nodeArtifactsDir);
+        const nodeId = path.basename(nodeArtifactsDir);
+        await writeDagNodeJsonArtifact(runDir, nodeId, "pi-extensions.json", {
+            schemaVersion: 1,
+            requested: [...requested],
+            resolved: resolved.map((entry) => ({
+                id: entry.id,
+                packageName: entry.packageName,
+                version: entry.version,
+                path: entry.entryPaths.join(", "),
+            })),
+            missing: missing.map((entry) => ({
+                id: entry.id,
+                reason: entry.reason,
+                ...(entry.detail ? { detail: entry.detail } : {}),
+            })),
+        });
+    }
+    catch {
+        // best-effort run fact; node execution continues
+    }
+}
 async function resolveValidatedFrontendBaseUrlFromContext(workspaceRoot, runDir) {
     try {
         const capability = JSON.parse(await readFile(path.join(runDir, "preflight-frontend-browser-tool-shell", "frontend-browser-capability.json"), "utf8"));
@@ -550,6 +581,29 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
                 };
             }
         }
+        // Plan 2026-08-21 D4/D6: resolve the node's explicit Pi extension
+        // allowlist against the settings inventory. Degrade on missing packages —
+        // never fail the node — and persist requested/resolved/missing facts.
+        const piExtensionsResolution = input.task.piExtensions
+            ? resolveDagPiExtensions({
+                repoRoot: input.cwd,
+                ids: input.task.piExtensions,
+            })
+            : undefined;
+        if (piExtensionsResolution) {
+            const piBackend = resolvePiBackend();
+            const cliDegraded = piBackend === "cli-only";
+            await writePiExtensionsArtifact(path.join(meta.runDir, input.task.id), input.task.piExtensions, cliDegraded ? [] : piExtensionsResolution.resolved, cliDegraded
+                ? input.task.piExtensions.map((id) => ({
+                    id,
+                    reason: "not-installed",
+                    detail: "cli backend does not load pi extensions (sdk-only v1)",
+                }))
+                : piExtensionsResolution.missing);
+        }
+        const piExtensionPaths = piExtensionsResolution && resolvePiBackend() !== "cli-only"
+            ? piExtensionsResolution.resolved.flatMap((entry) => entry.entryPaths)
+            : undefined;
         result = await piStepFn({
             attachedFiles: [],
             modelConfig: resolveDagPiModelConfig(input.model, input.thinking ? { thinking: input.thinking } : undefined),
@@ -564,6 +618,9 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
             stallTimeoutMs: input.stallTimeoutMs,
             abortGraceMs: input.abortGraceMs,
             ...(writerToolPolicy ? { writerToolPolicy } : {}),
+            ...(piExtensionPaths && piExtensionPaths.length > 0
+                ? { piExtensionPaths }
+                : {}),
             onActivity: bridgeActivity
                 ? (activity) => {
                     if (activity.kind === "lease" ||
@@ -885,22 +942,31 @@ export async function executeDagPiNode(input, meta, piStepFn = executePiStep, wr
     if (meta.writeGuardAttribution === "best-effort") {
         stderrParts.push("write guard note: concurrent rank writers use best-effort per-node attribution; keep same-rank writeSet entries disjoint");
     }
-    // Pure provider stall/timeout with zero workspace side effects is eligible
-    // for a single writer transport retry (implement-pi flake resilience).
-    // Partial writes, tool calls, or write-guard failures keep raw `timeout`.
-    const writerCleanTimeout = !mapped.ok &&
-        mapped.failureCategory === "timeout" &&
-        writeGuardOk &&
+    const rawFailureCategory = mapped.failureCategory;
+    const writeToolCallCount = readWriterThinkingExhaustionEvidence(result).writeToolCallCount;
+    const zeroSideEffects = writeGuardOk &&
         !completenessFailure &&
         !writerThinkingExhausted &&
-        isWriterTransportRetryCandidate(input.task) &&
+        !writerOutcomeViolation &&
         changeManifestChangedFiles !== undefined &&
         changeManifestChangedFiles.length === 0 &&
-        readWriterThinkingExhaustionEvidence(result).writeToolCallCount === 0;
+        writeToolCallCount === 0;
+    const targetCleanTransport = !mapped.ok &&
+        isTargetTemplateTransientRetryNode(input.task) &&
+        rawFailureCategory !== undefined &&
+        ["timeout", "network", "rate-limit", "unavailable"].includes(rawFailureCategory) &&
+        zeroSideEffects;
+    const frontendCleanTimeout = !mapped.ok &&
+        rawFailureCategory === "timeout" &&
+        isWriterTransportRetryCandidate(input.task) &&
+        !isTargetTemplateTransientRetryNode(input.task) &&
+        zeroSideEffects;
+    const writerCleanTimeout = targetCleanTransport || frontendCleanTimeout;
     return {
         ...mapped,
         ok: false,
         stderr: stderrParts.filter(Boolean).join("\n\n"),
+        ...(rawFailureCategory ? { rawFailureCategory } : {}),
         failureCategory: mapped.ok
             ? writeGuardOk
                 ? completenessFailure

package/dist/executors/pi-extension-resolver.js

@@ -0,0 +1,233 @@
+import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
+import os from "node:os";
+import path from "node:path";
+/**
+ * Pi extension allowlist for backend implementation DAGs (plan D2/D3).
+ * DAG specs carry these frozen short ids; the executor resolves them against
+ * the Pi agent settings package inventory at run time.
+ */
+export const DAG_PI_EXTENSION_IDS = ["pi-codegraph", "pi-lens"];
+/**
+ * Alias table: one DAG short id may resolve to multiple published package
+ * names. The first settings spec whose de-`npm:`-prefixed package name is in
+ * the alias set wins (project settings override user settings).
+ */
+export const DAG_PI_EXTENSION_PACKAGE_ALIASES = {
+    "pi-codegraph": [
+        "@estebanforge/pi-codegraph-enhanced",
+        "@vndv/pi-codegraph",
+        "pi-codegraph",
+    ],
+    "pi-lens": ["pi-lens"],
+};
+/**
+ * Extension tools that must never be activated as extras on any node (D5):
+ * they mutate files without the writer path policy. Writer nodes keep only
+ * policy-wrapped edit/write; read nodes keep only navigation tools.
+ */
+export const DAG_PI_EXTENSION_WRITE_TOOL_DENY = new Set([
+    "ast_grep_replace",
+]);
+/** Builtin tool names extensions must never replace (D5). */
+export const DAG_PI_EXTENSION_PROTECTED_BUILTIN_TOOLS = new Set(["read", "edit", "write", "bash", "grep", "find", "ls"]);
+/** Deny surface extensions must never activate (D5). */
+export const DAG_PI_EXTENSION_DENIED_TOOL_IDS = new Set([
+    "apply_patch",
+    "apply-patch",
+    "full-tools",
+    "full_tools",
+    "coding-chat",
+    "coding_chat",
+    "shell",
+]);
+export function isDagPiExtensionId(value) {
+    return DAG_PI_EXTENSION_IDS.includes(value);
+}
+function getPiAgentDir() {
+    const envDir = process.env.PI_CODING_AGENT_DIR;
+    if (envDir)
+        return envDir;
+    return path.join(os.homedir(), ".pi", "agent");
+}
+function readSettingsPackages(settingsPath) {
+    try {
+        if (!existsSync(settingsPath))
+            return undefined;
+        const parsed = JSON.parse(readFileSync(settingsPath, "utf8"));
+        if (!Array.isArray(parsed.packages))
+            return undefined;
+        return parsed.packages.filter((spec) => typeof spec === "string");
+    }
+    catch {
+        return undefined;
+    }
+}
+/** Strip an `npm:` prefix from a settings package spec, e.g. `npm:pi-lens`. */
+export function packageNameFromSpec(spec) {
+    return spec.startsWith("npm:") ? spec.slice(4) : spec;
+}
+function loadPackageInventories(repoRoot) {
+    const agentDir = getPiAgentDir();
+    const userSpecs = readSettingsPackages(path.join(agentDir, "settings.json"));
+    const projectSpecs = readSettingsPackages(path.join(repoRoot, ".pi", "settings.json"));
+    const inventories = [];
+    if (userSpecs) {
+        inventories.push({
+            specs: userSpecs,
+            nodeModulesRoot: path.join(agentDir, "npm", "node_modules"),
+        });
+    }
+    if (projectSpecs) {
+        inventories.push({
+            specs: projectSpecs,
+            nodeModulesRoot: path.join(repoRoot, "node_modules"),
+        });
+    }
+    return inventories;
+}
+const EXTENSION_FILE_RE = /\.(?:ts|js|mjs|cjs)$/;
+/**
+ * Resolve a package's `pi.extensions` entries to absolute entry paths.
+ * - A file entry maps to itself.
+ * - A directory entry expands to every extension file directly inside it
+ *   (e.g. `@estebanforge/pi-codegraph-enhanced` ships `./extensions` holding
+ *   `codegraph.ts`).
+ * No `pi.extensions` declaration, or a declaration whose paths do not exist,
+ * yields no entries (caller records `no-entry`) — no heuristic scanning
+ * outside declared paths.
+ */
+export function resolvePackageEntryPaths(packageDir) {
+    let manifest;
+    try {
+        manifest = JSON.parse(readFileSync(path.join(packageDir, "package.json"), "utf8"));
+    }
+    catch {
+        return { entryPaths: [], reason: "no-entry" };
+    }
+    const declared = manifest.pi?.extensions;
+    if (!Array.isArray(declared)) {
+        return { entryPaths: [], reason: "no-entry" };
+    }
+    const entryPaths = [];
+    for (const entry of declared) {
+        if (typeof entry !== "string" || entry.length === 0)
+            continue;
+        const resolved = path.resolve(packageDir, entry);
+        if (!existsSync(resolved))
+            continue;
+        let isDirectory = false;
+        try {
+            isDirectory = statSync(resolved).isDirectory();
+        }
+        catch {
+            continue;
+        }
+        if (!isDirectory) {
+            entryPaths.push(resolved);
+            continue;
+        }
+        let names;
+        try {
+            names = readdirSync(resolved);
+        }
+        catch {
+            continue;
+        }
+        for (const name of names.sort()) {
+            if (!EXTENSION_FILE_RE.test(name))
+                continue;
+            entryPaths.push(path.join(resolved, name));
+        }
+    }
+    if (entryPaths.length === 0) {
+        return { entryPaths: [], reason: "no-entry" };
+    }
+    return { entryPaths };
+}
+/**
+ * Resolve frozen DAG short ids against the Pi settings package inventory
+ * (user `~/.pi/agent/settings.json`, then project `.pi/settings.json` — a
+ * later match on the same alias wins). Pure synchronous read; never throws.
+ */
+export function resolveDagPiExtensions(input) {
+    const inventories = loadPackageInventories(input.repoRoot);
+    const resolved = [];
+    const missing = [];
+    for (const id of input.ids) {
+        if (!isDagPiExtensionId(id)) {
+            missing.push({
+                id,
+                reason: "package-unreadable",
+                detail: `unknown pi extension id: ${id}`,
+            });
+            continue;
+        }
+        const aliases = DAG_PI_EXTENSION_PACKAGE_ALIASES[id];
+        let matched;
+        // Later inventories (project) override earlier (user).
+        for (const inventory of inventories) {
+            for (const spec of inventory.specs) {
+                const packageName = packageNameFromSpec(spec);
+                if (!aliases.includes(packageName))
+                    continue;
+                const packageDir = path.join(inventory.nodeModulesRoot, ...packageName.split("/"));
+                if (!existsSync(packageDir))
+                    continue;
+                try {
+                    const manifest = JSON.parse(readFileSync(path.join(packageDir, "package.json"), "utf8"));
+                    if (typeof manifest.version !== "string")
+                        continue;
+                    matched = { packageDir, packageName, version: manifest.version };
+                }
+                catch {
+                    // unreadable manifest: skip this candidate
+                }
+            }
+        }
+        if (!matched) {
+            missing.push({ id, reason: "not-installed" });
+            continue;
+        }
+        const entries = resolvePackageEntryPaths(matched.packageDir);
+        if (entries.entryPaths.length === 0) {
+            missing.push({
+                id,
+                reason: entries.reason === "no-entry" ? "no-entry" : "package-unreadable",
+            });
+            continue;
+        }
+        resolved.push({
+            id,
+            packageName: matched.packageName,
+            version: matched.version,
+            entryPaths: entries.entryPaths,
+        });
+    }
+    return { resolved, missing };
+}
+/**
+ * Compute extension tool names that may join a node's active tool allowlist
+ * after session creation (D5): new tool names registered by the loaded
+ * extensions, minus protected builtins, the deny surface, and write-capable
+ * extension tools (both buckets — writer writes stay policy-wrapped only).
+ */
+export function computeExtensionToolAllowlistExtras(sessionTools, baseline) {
+    const baselineLower = new Set(baseline.map((name) => name.toLowerCase()));
+    const extras = [];
+    for (const tool of sessionTools) {
+        const name = tool && typeof tool.name === "string" ? tool.name : undefined;
+        if (!name)
+            continue;
+        const lower = name.toLowerCase();
+        if (baselineLower.has(lower))
+            continue;
+        if (DAG_PI_EXTENSION_PROTECTED_BUILTIN_TOOLS.has(lower))
+            continue;
+        if (DAG_PI_EXTENSION_DENIED_TOOL_IDS.has(lower))
+            continue;
+        if (DAG_PI_EXTENSION_WRITE_TOOL_DENY.has(lower))
+            continue;
+        extras.push(name);
+    }
+    return extras;
+}

package/dist/executors/pi-sdk-executor.js

@@ -1,5 +1,6 @@
 import { appendFile, mkdir } from "node:fs/promises";
 import path from "node:path";
+import { computeExtensionToolAllowlistExtras } from "./pi-extension-resolver.js";
 import { BoundedTextPreview, classifyPiFailure, createPiJsonlStreamCollector, DEFAULT_ABORT_GRACE_MS, DEFAULT_STALL_TIMEOUT_MS, DEFAULT_TIMEOUT_MS, extractPiEventStopReason, extractAssistantTextFromPiJson, } from "./pi-executor.js";
 import { serializeSessionEvent } from "./pi-event-serializer.js";
 import { PI_RECOMMENDED_COMPACTION, PI_RECOMMENDED_RETRY, } from "../shared/pi-retry-settings.js";
@@ -229,6 +230,9 @@ async function createSdkSession(sdk, input, shared) {
         noContextFiles: true,
         noSkills: true,
         noExtensions: true,
+        ...(input.additionalExtensionPaths && input.additionalExtensionPaths.length > 0
+            ? { additionalExtensionPaths: input.additionalExtensionPaths }
+            : {}),
         ...(settingsManager ? { settingsManager } : {}),
         appendSystemPromptOverride: (base) => [
             ...base,
@@ -236,6 +240,19 @@ async function createSdkSession(sdk, input, shared) {
         ],
     });
     await loader.reload();
+    // Discover extension-registered tool names BEFORE session creation (plan
+    // 2026-08-21 D5): verified against the real SDK — passing extension tool
+    // names in createAgentSession `tools` activates them, while a post-create
+    // setActiveToolsByName pin leaves the active set unchanged. Merge filtered
+    // extras into the create-time allowlist; protected builtins, the deny
+    // surface, and write-capable extension tools never join.
+    const extensionToolExtras = [];
+    if (input.additionalExtensionPaths &&
+        input.additionalExtensionPaths.length > 0) {
+        const registered = collectLoaderExtensionToolNames(loader);
+        extensionToolExtras.push(...computeExtensionToolAllowlistExtras(registered.map((name) => ({ name })), input.toolNames));
+    }
+    const sessionToolNames = [...input.toolNames, ...extensionToolExtras];
     if (input.requireWriterCustomTools) {
         if (!Array.isArray(input.customTools) || input.customTools.length === 0) {
             throw new Error("pi writer tool policy missing customTools; refusing to create uncontrolled writer session");
@@ -245,7 +262,7 @@ async function createSdkSession(sdk, input, shared) {
         cwd: input.cwd,
         sessionManager: SessionManager.inMemory(input.cwd),
         resourceLoader: loader,
-        tools: input.toolNames,
+        tools: sessionToolNames,
         modelRuntime,
         ...(settingsManager ? { settingsManager } : {}),
         ...(model ? { model } : {}),
@@ -254,6 +271,22 @@ async function createSdkSession(sdk, input, shared) {
             ? { customTools: input.customTools }
             : {}),
     });
+    if (extensionToolExtras.length > 0) {
+        // Fail closed when merged extension tools did not activate: a silent
+        // pin failure must never let the node run without its declared tools.
+        const activeTools = created.session.agent?.state?.tools;
+        if (!Array.isArray(activeTools)) {
+            throw new Error("pi SDK session does not expose agent.state.tools; refusing unverifiable extension-tool activation");
+        }
+        const activeNames = new Set(activeTools
+            .map((tool) => typeof tool?.name === "string" ? tool.name : undefined)
+            .filter((name) => Boolean(name)));
+        for (const extensionToolName of extensionToolExtras) {
+            if (!activeNames.has(extensionToolName)) {
+                throw new Error(`pi SDK extension-tool activation mismatch for ${extensionToolName}: tool did not activate at session creation`);
+            }
+        }
+    }
     const customToolNames = (input.customTools ?? [])
         .map((tool) => isRecord(tool) && typeof tool.name === "string" ? tool.name : undefined)
         .filter((name) => Boolean(name));
@@ -266,7 +299,7 @@ async function createSdkSession(sdk, input, shared) {
             .map((tool) => typeof tool?.name === "string" ? tool.name : undefined)
             .filter((name) => Boolean(name)));
         for (const customToolName of customToolNames) {
-            const expectedActive = input.toolNames.includes(customToolName);
+            const expectedActive = sessionToolNames.includes(customToolName);
             if (activeNames.has(customToolName) !== expectedActive) {
                 throw new Error(`pi SDK custom-tool activation mismatch for ${customToolName}: expectedActive=${expectedActive}`);
             }
@@ -274,6 +307,47 @@ async function createSdkSession(sdk, input, shared) {
     }
     return created.session;
 }
+/**
+ * Collect tool names registered by loader-resolved extensions. The SDK exposes
+ * each extension's `tools` as a Record (name → definition) or Map; both shapes
+ * are handled. Missing getExtensions (older SDK) yields no extras — the node
+ * then runs with the baseline allowlist only.
+ */
+function collectLoaderExtensionToolNames(loader) {
+    const getExtensions = loader?.getExtensions;
+    if (typeof getExtensions !== "function")
+        return [];
+    let parsed;
+    try {
+        parsed = getExtensions.call(loader);
+    }
+    catch {
+        return [];
+    }
+    if (!parsed || typeof parsed !== "object")
+        return [];
+    const extensions = parsed.extensions;
+    if (!Array.isArray(extensions))
+        return [];
+    const names = [];
+    for (const extension of extensions) {
+        if (!extension || typeof extension !== "object")
+            continue;
+        const tools = extension.tools;
+        if (tools instanceof Map) {
+            for (const key of tools.keys()) {
+                if (typeof key === "string")
+                    names.push(key);
+            }
+        }
+        else if (tools && typeof tools === "object") {
+            for (const key of Object.keys(tools)) {
+                names.push(key);
+            }
+        }
+    }
+    return names;
+}
 /**
  * Throttle high-noise Pi SDK events from session-events.jsonl.
  * Always keeps tool lifecycle, turn/agent end, assistant messages, and errors.
@@ -545,6 +619,9 @@ export async function executeSingleSdkAttempt(options) {
                 ? { customTools: writerCustomTools }
                 : {}),
             ...(requireWriterCustomTools ? { requireWriterCustomTools: true } : {}),
+            ...(options.piExtensionPaths && options.piExtensionPaths.length > 0
+                ? { additionalExtensionPaths: options.piExtensionPaths }
+                : {}),
         });
         let resolveStall;
         let resolveSettlement;

package/dist/executors/shell-executor.js

@@ -123,6 +123,39 @@ function shouldReuseSuccessfulShellCommands(shell) {
 const DEFAULT_SHELL_TIMEOUT_MS = 300_000;
 const SUMMARY_STDOUT_MAX = 4_000;
 const SUMMARY_STDERR_MAX = 2_000;
+const TEST_TIMEOUT_CLASSIFY_MAX_CHARS = 8_000;
+const TEST_TIMEOUT_MARKERS = [
+    /Test timed out in \d+ms/i,
+    /Timeout of \d+ms exceeded/i,
+];
+const DETERMINISTIC_NON_TIMEOUT_FAILURE_MARKERS = [
+    /AssertionError/i,
+    /expected:?\s/i,
+    /received:?\s/i,
+    /Snapshot /i,
+    /toMatchSnapshot/i,
+    /SyntaxError/i,
+    /TS\d{4}/,
+    /error TS/i,
+    /eslint/i,
+    /compile error/i,
+    /Failed to compile/i,
+];
+function isIdentifiedTestCommand(command) {
+    return /(?:^|[\s"'`])(?:npx\s+)?vitest(?:\s|$)|(?:^|[\s"'`])npm(?:\.cmd)?\s+test(?:\s|$)|(?:^|[\s"'`])pnpm(?:\.cmd)?\s+test(?:\s|$)|(?:^|[\s"'`])yarn(?:\.cmd)?\s+test(?:\s|$)/i.test(command);
+}
+function classifyShellCommandFailure(input) {
+    if (input.processCategory !== "nonzero-exit")
+        return input.processCategory;
+    if (!isIdentifiedTestCommand(input.command))
+        return "nonzero-exit";
+    const sample = `${input.stdout}\n${input.stderr}`.slice(0, TEST_TIMEOUT_CLASSIFY_MAX_CHARS);
+    const hasTimeoutMarker = TEST_TIMEOUT_MARKERS.some((marker) => marker.test(sample));
+    if (!hasTimeoutMarker)
+        return "nonzero-exit";
+    const mixedDeterministicFailure = DETERMINISTIC_NON_TIMEOUT_FAILURE_MARKERS.some((marker) => marker.test(sample));
+    return mixedDeterministicFailure ? "nonzero-exit" : "test-timeout";
+}
 const NODE_STDOUT_MAX = 12_000;
 const NODE_STDERR_MAX = 8_000;
 const MAX_BUFFERED_OUTPUT_BYTES = 64 * 1024;
@@ -329,16 +362,21 @@ export async function executeShellCommand(input) {
             });
         });
         child.on("close", (code) => {
-            let failureCategory = "nonzero-exit";
+            let processCategory = "nonzero-exit";
             if (timedOut) {
-                failureCategory = "timeout";
+                processCategory = "timeout";
             }
             else if (code === 0) {
-                failureCategory = "success";
+                processCategory = "success";
             }
             finish({
                 exitCode: code,
-                failureCategory,
+                failureCategory: classifyShellCommandFailure({
+                    command: input.command,
+                    processCategory,
+                    stdout: formatBoundedOutput(stdout),
+                    stderr: formatBoundedOutput(stderr),
+                }),
                 ok: code === 0 && !timedOut,
             });
         });
@@ -2878,6 +2916,7 @@ export async function executeDagShellNode(input, meta) {
     const cwd = resolveShellCwd(input.cwd, shell.cwd);
     const timeoutMs = shell.timeoutMs ?? DEFAULT_SHELL_TIMEOUT_MS;
     const failFast = shell.failFast !== false;
+    const attemptNumber = input.attempt ?? 1;
     const results = [];
     const executedCommandKeys = new Set();
     let beforeStatus;
@@ -2973,8 +3012,8 @@ export async function executeDagShellNode(input, meta) {
             dagRunMeta: { runDir: meta.runDir, runId: meta.runId },
             reportActivity: input.reportActivity,
             outputArtifacts: {
-                stdoutPath: path.join(meta.runDir, input.task.id, "commands", `${commandNumber}.stdout.txt`),
-                stderrPath: path.join(meta.runDir, input.task.id, "commands", `${commandNumber}.stderr.txt`),
+                stdoutPath: path.join(meta.runDir, input.task.id, `attempt-${attemptNumber}`, "commands", `${commandNumber}.stdout.txt`),
+                stderrPath: path.join(meta.runDir, input.task.id, `attempt-${attemptNumber}`, "commands", `${commandNumber}.stderr.txt`),
             },
         });
         results.push(result);
@@ -3033,6 +3072,7 @@ export async function executeDagShellNode(input, meta) {
         schemaVersion: 1,
         runId: meta.runId,
         nodeId: input.task.id,
+        attempt: attemptNumber,
         workspaceFingerprint: workspaceFingerprint ?? null,
         failFast,
         commands: results.map((result) => ({
@@ -3051,6 +3091,7 @@ export async function executeDagShellNode(input, meta) {
             subchecks: parseCheckRepoSubcheckDurations(result),
         })),
     };
+    await writeDagNodeJsonArtifact(meta.runDir, input.task.id, path.posix.join(`attempt-${attemptNumber}`, "shell-command-receipt.json"), receipt);
     await writeDagNodeJsonArtifact(meta.runDir, input.task.id, "shell-command-receipt.json", receipt);
     const summaryMarkdown = buildShellResultSummaryMarkdown({
         nodeId: input.task.id,
@@ -3102,6 +3143,9 @@ export async function executeDagShellNode(input, meta) {
         stdout: `${recordPrefix}${reuseNote}${aggregate.stdout}`,
         stderr: stderrParts.filter(Boolean).join("\n\n"),
         failureCategory,
+        ...(firstFailure?.failureCategory
+            ? { rawFailureCategory: firstFailure.failureCategory }
+            : {}),
         durationMs: Date.now() - started,
         // extra field consumed by node-execution if supported; also embedded in stdout marker
         ...{ commandResults },

package/dist/executors/shell-write-guard.js

@@ -98,6 +98,13 @@ export function isEphemeralToolCachePath(filePath) {
     const normalized = normalizePath(filePath);
     if (!normalized)
         return false;
+    // CodeGraph extension sync side effect (plan 2026-08-21 D7): a loaded
+    // pi-codegraph extension keeps an existing .codegraph/ index fresh at
+    // session start; those index files are extension-owned run state, not
+    // writer output, so they never count as write-guard violations.
+    if (normalized === ".codegraph" || normalized.startsWith(".codegraph/")) {
+        return true;
+    }
     if (normalized === ".pytest_cache" ||
         normalized.startsWith(".pytest_cache/")) {
         return true;

package/dist/shared/operator/capabilities.js

@@ -1822,6 +1822,13 @@ export const OPERATOR_COMMAND_COVERAGE = Object.freeze([
         source: "loop-agent",
         exclusionReason: "Not exposed to Operator Chat in M5.",
     },
+    {
+        command: "loop-agent init model-catalog",
+        coverage: "excluded",
+        action: null,
+        source: "loop-agent",
+        exclusionReason: "Not exposed to Operator Chat in M5.",
+    },
     {
         command: "loop-agent init doctor",
         coverage: "excluded",