@agent-native/core 0.95.1 → 0.97.0

package/dist/server/agent-chat-plugin.js

@@ -1,7 +1,7 @@
 import nodePath from "node:path";
-import { createError, defineEventHandler, setResponseStatus, setResponseHeader, getMethod, getQuery, getHeader, } from "h3";
+import { createError, defineEventHandler, setResponseStatus, setResponseHeader, getMethod, getQuery, getHeader, getRequestIP, } from "h3";
 import { appendA2AArtifactLinks, buildA2ARecoverableArtifactMessage, } from "../a2a/artifact-response.js";
-import { hasConfiguredA2ASecret, isA2AProductionRuntime, } from "../a2a/auth-policy.js";
+import { hasConfiguredA2ASecret, isLoopbackAddress, isTrustedLocalRuntime, } from "../a2a/auth-policy.js";
 import { applyAgentTextEventToBuffer, collectFinalResponseTextFromAgentEvents, } from "../a2a/response-text.js";
 import { updateTaskStatusMessage } from "../a2a/task-store.js";
 import { ACTION_CHAT_UI_DATA_WIDGET_RENDERER } from "../action-ui.js";
@@ -9,7 +9,7 @@ import { canUpdateAgentAppModelDefaultSettings, normalizeAgentAppModelDefaultApp
 import { DEFAULT_ANTHROPIC_MODEL } from "../agent/default-model.js";
 import { AGENT_CHAT_BACKGROUND_RUN_FIELD, AGENT_CHAT_PROCESS_RUN_PATH, backgroundRunMarkerExpectsBackgroundRuntime, isInBackgroundFunctionRuntime, prepareProcessRunRequest, } from "../agent/durable-background.js";
 import { resolveEngine, createAnthropicEngine, getStoredModelForEngine, normalizeModelForEngine, getAgentEngineEntry, isAgentEnginePackageInstalled, isStoredEngineUsableForRequest, listAgentEngines, registerBuiltinEngines, } from "../agent/engine/index.js";
-import { createProductionAgentHandler, actionsToEngineTools, executeAgentToolCall, getActiveRunForThreadAsync, abortRunDurably, subscribeToRun, } from "../agent/production-agent.js";
+import { createProductionAgentHandler, actionsToEngineTools, executeAgentToolCall, filterInitialEngineTools, getActiveRunForThreadAsync, abortRunDurably, subscribeToRun, } from "../agent/production-agent.js";
 import { runAgentLoopDirectWithSoftTimeout } from "../agent/run-loop-with-resume.js";
 import { callerHasRunAccess, callerHasThreadAccess, } from "../agent/run-ownership.js";
 import { CLAIMED_BACKGROUND_WORKER_FAILED_ERROR_EVENT, ensureTerminalRunEvent, readBackgroundRunClaim, recordRunDiagnostic, RUN_DIAG_STAGE, setRunError, setRunTerminalReason, updateRunStatusIfRunning, } from "../agent/run-store.js";
@@ -727,6 +727,9 @@ function appStateKeyForBrowserTab(key, browserTabId) {
  * entries are pruned on read.
  */
 const generateTitleRateLimit = new Map();
+/** Only sweep drained rate-limit entries once the map grows past this size,
+ * so the common small-map case stays O(1). */
+const RATE_LIMIT_SWEEP_THRESHOLD = 1000;
 /**
  * Creates the `set-search-params` / `set-url-path` tools. Writes a one-shot
  * URL command to application_state; the client's URLSync component applies
@@ -1412,6 +1415,10 @@ async function createChatScriptEntries() {
                         description: "Output format",
                         enum: ["json", "text"],
                     },
+                    includeArchived: {
+                        type: "boolean",
+                        description: "Also include archived chats in the results. Archived chats are excluded by default.",
+                    },
                 },
             },
         }, searchMod.default);
@@ -1453,6 +1460,10 @@ async function createChatScriptEntries() {
                                 description: "(search) Output format",
                                 enum: ["json", "text"],
                             },
+                            includeArchived: {
+                                type: "boolean",
+                                description: "(search) Also include archived chats in the results. Archived chats are excluded by default, matching the 'archive' action's effect on the chat list.",
+                            },
                             id: {
                                 type: "string",
                                 description: "(open, rename, pin, unpin, archive) The chat thread ID to manage",
@@ -1995,6 +2006,18 @@ export function runA2AAgentLoop(runOptions, pluginOptions, timeoutOptions, runne
         finalResponseGuard: pluginOptions.finalResponseGuard,
     }, pluginOptions.runSoftTimeoutMs, timeoutOptions);
 }
+/**
+ * Keep delegated A2A turns on the same compact initial tool surface as
+ * interactive chat. `tool-search` remains in the initial set, while the
+ * complete registry is supplied separately so the run loop can load a matched
+ * schema after a tool-search result.
+ */
+export function createA2AEngineToolSurface(availableTools, initialToolNames) {
+    return {
+        tools: filterInitialEngineTools(availableTools, initialToolNames),
+        availableTools,
+    };
+}
 /**
  * Verbose framework sections returned by the `get-framework-context` tool.
  * Keyed by topic so the agent can request specific sections.
@@ -2025,9 +2048,9 @@ Use for charts, visualizations, previews. Don't use for simple text/tables or ex
     "chat-history": `### Chat History
 
 You can search and restore previous chat conversations using \`chat-history\`:
-- \`chat-history\` (action: "search") — Search or list past chat threads by keyword
+- \`chat-history\` (action: "search") — Search or list past chat threads by keyword. Archived threads are excluded by default; pass \`includeArchived: true\` to also see them.
 - \`chat-history\` (action: "open") — Open a chat thread in the UI as a new tab and focus it
-- \`chat-history\` (actions: "rename", "pin", "unpin", "archive") — Organize a known chat thread by ID
+- \`chat-history\` (actions: "rename", "pin", "unpin", "archive") — Organize a known chat thread by ID. Archiving a thread hides it from the default chat list and search.
 
 When the user asks to find a previous conversation, use \`chat-history\` with action "search" first to find matching threads, then action "open" to restore the one they want.`,
     "agent-teams": `### Agent Teams — Orchestration
@@ -3785,7 +3808,7 @@ export function createAgentChatPlugin(options) {
                             ...(resolvedProdCodeExec !== "off" ? runCodeTool : {}),
                             ...prodCodingTools,
                         });
-                    const a2aTools = actionsToEngineTools(a2aActions);
+                    const a2aToolSurface = createA2AEngineToolSurface(actionsToEngineTools(a2aActions), options?.initialToolNames);
                     // Precise current time rides the user message (not the cached
                     // system-prompt prefix) — same pattern as the interactive handler.
                     const a2aMessages = [
@@ -3802,12 +3825,13 @@ export function createAgentChatPlugin(options) {
                     const a2aToolResults = [];
                     let lastRecoverableArtifactText = "";
                     const controller = new AbortController();
-                    console.log(`[A2A] Starting agent loop: ${a2aTools.length} tools, prompt ${systemPrompt.length} chars`);
+                    console.log(`[A2A] Starting agent loop: ${a2aToolSurface.tools.length}/${a2aToolSurface.availableTools.length} initial tools, prompt ${systemPrompt.length} chars`);
                     await runA2AAgentLoop({
                         engine: a2aEngine,
                         model,
                         systemPrompt,
-                        tools: a2aTools,
+                        tools: a2aToolSurface.tools,
+                        availableTools: a2aToolSurface.availableTools,
                         messages: a2aMessages,
                         actions: a2aActions,
                         send: (event) => {
@@ -4991,11 +5015,14 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                         return { error: "Invalid or expired processor token" };
                     }
                 }
-                else if (isA2AProductionRuntime()) {
-                    setResponseStatus(event, 503);
-                    return {
-                        error: "Agent Teams processor not configured — set A2A_SECRET on this deployment.",
-                    };
+                else {
+                    const loopback = isLoopbackAddress(getRequestIP(event, { xForwardedFor: false }));
+                    if (!isTrustedLocalRuntime({ loopback })) {
+                        setResponseStatus(event, 503);
+                        return {
+                            error: "Agent Teams processor not configured — set A2A_SECRET on this deployment (or A2A_ALLOW_UNSIGNED_INTERNAL=1 for trusted local dev).",
+                        };
+                    }
                 }
                 try {
                     return await processAgentTeamRun({
@@ -5523,6 +5550,10 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                 // Stream NDJSON — each source flushes its batch as soon as it's ready.
                 setResponseHeader(event, "Content-Type", "application/x-ndjson");
                 setResponseHeader(event, "Cache-Control", "no-cache");
+                // Lets `cancel()` signal in-flight source work to stop early on
+                // client disconnect, instead of only being noticed the next time
+                // `flush()`'s `controller.enqueue` throws.
+                const mentionsAbort = new AbortController();
                 const stream = new ReadableStream({
                     start(controller) {
                         return runWithRequestContext({
@@ -5531,17 +5562,21 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                         }, () => mentionsStreamWork(controller));
                     },
                     cancel() {
-                        // Client disconnected — stop enqueuing
+                        // Client disconnected — stop enqueuing and let in-flight source
+                        // work short-circuit instead of running to completion unseen.
+                        mentionsAbort.abort();
                     },
                 });
                 return stream;
                 async function mentionsStreamWork(controller) {
                     const MAX_RESULTS = 50;
                     let totalSent = 0;
-                    let cancelled = false;
+                    let cancelled = mentionsAbort.signal.aborted;
                     const flush = (batch) => {
-                        if (cancelled)
+                        if (cancelled || mentionsAbort.signal.aborted) {
+                            cancelled = true;
                             return;
+                        }
                         const filtered = batch.filter(matchesQuery);
                         if (filtered.length === 0)
                             return;
@@ -5607,6 +5642,9 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                     }
                     // 3. Custom mention providers (each flushes independently)
                     for (const [key, provider] of Object.entries(mentionProviders)) {
+                        // Client already disconnected — don't spawn more provider work.
+                        if (mentionsAbort.signal.aborted)
+                            break;
                         sources.push((async () => {
                             try {
                                 const providerItems = await provider.search(q, event);
@@ -5675,8 +5713,9 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                         }
                     })());
                     await Promise.all(sources);
-                    if (!cancelled)
+                    if (!cancelled && !mentionsAbort.signal.aborted) {
                         controller.close();
+                    }
                 }
             }));
             // ─── Generate thread title ──────────────────────────────────────────
@@ -5699,6 +5738,19 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                 }
                 recent.push(now);
                 generateTitleRateLimit.set(ownerEmail, recent);
+                // Opportunistic eviction: keep the map from growing unbounded by
+                // the count of distinct users on a long-lived process. Drop any
+                // key whose window has fully drained (no timestamps within the
+                // rolling window).
+                if (generateTitleRateLimit.size > RATE_LIMIT_SWEEP_THRESHOLD) {
+                    for (const [email, times] of generateTitleRateLimit) {
+                        if (email === ownerEmail)
+                            continue;
+                        if (times.every((t) => now - t >= limitWindowMs)) {
+                            generateTitleRateLimit.delete(email);
+                        }
+                    }
+                }
                 const body = await readBody(event);
                 const message = body?.message;
                 if (!message || typeof message !== "string") {
@@ -6510,8 +6562,11 @@ Non-code requests are still fine on this surface: read data, navigate the UI, su
                 }
                 // Validate + HMAC-authenticate the self-dispatch and prepare the
                 // background-worker body. Pure decision (unit-tested in
-                // durable-background.spec.ts); the route only wires it to h3.
-                const prepared = prepareProcessRunRequest(processBody, getHeader(event, "authorization"));
+                // durable-background.spec.ts); the route only wires it to h3. This
+                // handler DOES have the h3 event, so it can see the real socket
+                // peer — thread the loopback signal through so unsigned local-dev
+                // self-dispatch (no A2A_SECRET) still works over 127.0.0.1/::1.
+                const prepared = prepareProcessRunRequest(processBody, getHeader(event, "authorization"), isLoopbackAddress(getRequestIP(event, { xForwardedFor: false })));
                 if (!prepared.ok) {
                     // DIAGNOSTIC: record the auth/validation failure ONTO the run
                     // before returning the error status. Without this, a 401 (e.g.