@staix/agent-hub 0.12.7 → 0.12.9

package/src/models/route/text.ts

@@ -0,0 +1,61 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+// Ported to TypeScript from NVIDIA NeMo Switchyard crates/libsy/src/algorithms/advisor_gate/transcript.rs and algorithms/util/{escalation,prompts,llm_judge}.rs at commit c8848511, modified.
+
+import type { ChatMessage } from "../../omniroute/client.ts";
+
+/** Unicode code-point aware middle truncation used by Switchyard's advisor. */
+export function middleDrop(text: string, cap: number): string {
+  const chars = Array.from(text);
+  if (chars.length <= cap) return text;
+  const headCount = Math.max(0, Math.floor(cap / 4));
+  const tailCount = Math.max(0, cap - headCount);
+  return `${chars.slice(0, headCount).join("")}\n...<middle of the conversation truncated>...\n${chars.slice(-tailCount || chars.length).join("")}`;
+}
+
+/** Bounded head/tail truncation for escalation summaries, measured in code points. */
+export function truncateCodepoints(text: string, limit: number): string {
+  const chars = Array.from(text);
+  if (chars.length <= limit) return text;
+  const marker = " ...[trimmed] ";
+  const keep = Math.min(chars.length, Math.max(20, limit - Array.from(marker).length));
+  const head = Math.floor(keep * 2 / 3);
+  const tail = keep - head;
+  return `${chars.slice(0, head).join("")}${marker}${chars.slice(-tail).join("")}`;
+}
+
+const verdictPattern = /^[\s*_#>"'(\[`]*(?:(?:final\s+)?verdict\s*:\s*[\s*_#>"'(\[`]*)?(APPROVE|REDO)(?![\p{Alphabetic}\p{M}\p{Nd}\p{Pc}\p{Join_Control}])/iu;
+
+export type AdvisorVerdict = { kind: "approve" } | { kind: "redo"; plan: string };
+
+/** Only an anchored first-word verdict is trusted. */
+export function parseAdvisorVerdict(reply: string): AdvisorVerdict | undefined {
+  const trimmed = reply.trim();
+  const match = verdictPattern.exec(trimmed);
+  if (!match) return undefined;
+  if (match[1]?.toUpperCase() === "APPROVE") return { kind: "approve" };
+  const tail = trimmed.slice(match[0].length).trimStart().replace(/^[ *_:\n-]+/u, "").trim();
+  return { kind: "redo", plan: tail || trimmed };
+}
+
+/** Removes only the JSON fences supported by Switchyard's typed judge parser. */
+export function stripJsonFence(text: string): string {
+  let value = text.trim();
+  if (!value.startsWith("```")) return value;
+  value = value.slice(3);
+  if (value.startsWith("json")) value = value.slice(4);
+  value = value.replace(/^[\n\r]+/u, "");
+  return value.endsWith("```") ? value.slice(0, -3).trim() : value;
+}
+
+/** Add a routing note without creating consecutive user turns. */
+export function appendNote(messages: readonly ChatMessage[], note: string): ChatMessage[] {
+  const output = messages.map(message => ({ ...message }));
+  const last = output.at(-1);
+  if (last?.role === "user") {
+    last.content = last.content ? `${last.content}\n${note}` : note;
+  } else {
+    output.push({ role: "user", content: note });
+  }
+  return output;
+}

package/src/omniroute/client.ts

@@ -24,8 +24,12 @@ export interface ChatMessage {
   content: string | null;
   tool_calls?: ToolCall[];
   tool_call_id?: string;
+  /** Internal local-tool outcome, consumed by route normalization and omitted from provider transport. */
+  is_error?: boolean;
 }
 export interface ChatResult {
+  /** Internal routing telemetry id, never part of a provider payload. */
+  routeDecision?: string;
   message: ChatMessage;
   /** Usage counters returned by the provider, when valid. No missing counter is inferred as zero. */
   usage?: NormalizedUsage;
@@ -37,6 +41,8 @@ export interface ChatResult {
   selectedModel?: string;
 }
 export interface ChatOptions {
+  /** Refuse a changed/off-campus gateway immediately before transport. */
+  onCampusOnly?: boolean;
   signal?: AbortSignal;
   /** Switchyard base (`http://127.0.0.1:<port>/v1`). The sidecar holds the key, so none is sent. */
   via?: string;
@@ -155,6 +161,7 @@ export class OmniRoute {
   async chat(body: { model: string; messages: ChatMessage[]; tools?: unknown[]; max_tokens?: number }, opts: ChatOptions = {}): Promise<ChatResult> {
     const base = opts.via ?? (await this.base());
     if (!base) throw new Error("no model gateway is configured or reachable (omniroute.urls in .agenthub/config.json; see ahub doctor)");
+    if (opts.onCampusOnly && (opts.via || this.isAccessHost(base))) throw new Error("PII model calls require the confirmed campus gateway");
     const headers: Record<string, string> = { "content-type": "application/json" };
     if (opts.via) {
       if (opts.sessionId) headers["x-switchyard-session-id"] = opts.sessionId;
@@ -168,7 +175,7 @@ export class OmniRoute {
       res = await fetch(`${base}/chat/completions`, {
         method: "POST",
         headers,
-        body: JSON.stringify({ ...body, stream: false }),
+        body: JSON.stringify({ ...body, messages: body.messages.map(({ is_error: _internalError, ...message }) => message), stream: false }),
         ...(opts.signal ? { signal: opts.signal } : {}),
       });
     } catch (e) {

package/templates/routing.toml

@@ -93,3 +93,31 @@ peers = ["pi", "local", "kimi"]
 route = "sy/fast"
 pi_backend = "mlx"
 escalate_to = ["codex", "kimi"]
+
+# In-process L2 routes are opt-in: set local.route or a class route to hub/<id>.
+# They never enter the optional Switchyard sidecar TOML. Defaults are efficient fast, capable coding.
+[hub_routes."hub/stage"]
+type = "stage"
+efficient = "fast"
+capable = "coding"
+confidence_threshold = 0.5
+hold_turns = 2
+
+[hub_routes."hub/plan"]
+type = "plan_execute"
+efficient = "fast"
+capable = "coding"
+
+[hub_routes."hub/advisor"]
+type = "advisor"
+efficient = "fast"
+capable = "coding"
+max_reviews = 1
+trigger = "no_tool_call"
+gate_min_tool_results = 1
+
+[hub_routes."hub/escalation"]
+type = "escalation"
+efficient = "fast"
+capable = "coding"
+confirmations = 2