sample: add ticket-triage-agent (System 1/System 2 pattern with Jev)

packages/uipath/samples/ticket-triage-agent/.env.example

@@ -0,0 +1,9 @@
+UIPATH_URL=https://cloud.uipath.com/your_org/your_tenant
+UIPATH_ACCESS_TOKEN=your_access_token_here
+
+# Required for the escalation path: Action Center tasks must be created in
+# an Orchestrator folder.
+UIPATH_FOLDER_PATH=Shared
+
+# Required for Tier 1 triage: API key for TypeSafe AI's Jev model.
+TYPESAFE_API_KEY=your_typesafe_api_key_here

packages/uipath/samples/ticket-triage-agent/README.md

@@ -0,0 +1,155 @@
+# Ticket Triage Agent: System 1 / System 2 Pattern
+
+This sample demonstrates a two-tier triage pattern for support tickets:
+
+1. **Tier 1 (System 1) - fast structured triage.** A "System One" model
+   takes the ticket text and a set of typed questions, and returns
+   calibrated typed answers (a classification, a 0-1 probability, and a
+   score) instead of free text - orders of magnitude cheaper and faster than an
+   LLM call. This sample uses TypeSafe AI's `Jev` model via the
+   `typesafe-sdk` package.
+2. **Tier 2 (System 2) - branch on the triage result.**
+   - If the ticket is urgent, the customer sounds frustrated, or the
+     department routing is low-confidence, the agent **escalates to a
+     human** via a UiPath Action Center QuickForm task, attaching the Tier 1
+     decision as context so the reviewer isn't starting cold.
+   - Otherwise, the agent calls a **real LLM** (UiPath LLM Gateway) with a
+     department-specific system prompt to draft a reply, and returns it as
+     an auto-resolved ticket. No human, and no expensive LLM call for
+     routing, is needed for the common case.
+
+## About the "Jev" model (important - read this)
+
+This sample calls TypeSafe AI's real `typesafe-sdk` PyPI package, not a
+mock. `typesafe-sdk` is a young package (first published 2026-09-09), so
+before wiring it in we statically inspected the wheel's source (no
+install/execution): it's a normal, apparently auto-generated API client
+(the response schemas reference an OpenAPI spec) with no `eval`/`exec`/
+`subprocess` calls, and no exfiltration of environment variables. Its
+`httpx2` dependency turned out to be an unrelated, independent package
+(the pydantic org's next-generation `httpx`), not something specific to
+TypeSafe AI. If you're pulling a new AI SDK into your own project, reading
+the source before trusting it is a cheap, worthwhile step regardless of how
+long the package has existed.
+
+```python
+from typesafe_sdk import Choice, Noul, Score, TypeSafeClient
+
+client = TypeSafeClient()  # reads TYPESAFE_API_KEY from the environment
+response = client.system_one(state=..., questions={...})
+```
+
+`pyproject.toml` pins `typesafe-sdk>=0.7.0` rather than leaving it
+unbounded, since `uipath`'s own `uv` install already applies a
+minimum-package-age safety check that skips the very newest release.
+
+## Prerequisites
+
+* [UV package manager](https://docs.astral.sh/uv/) installed
+* A TypeSafe AI API key (`TYPESAFE_API_KEY`) for the Tier 1 triage step
+* A UiPath Orchestrator tenant with:
+  * Access to the LLM Gateway (for the auto-reply path)
+  * Access to Action Center (for the escalation path)
+
+## Setup
+
+### Step 1: Create and activate a virtual environment
+
+```bash
+uv venv
+source .venv/bin/activate  # Windows: .venv\Scripts\activate
+```
+
+### Step 2: Install dependencies
+
+```bash
+uv sync
+```
+
+### Step 3: Configure credentials
+
+Copy `.env.example` to `.env` and fill in your Orchestrator URL, access
+token, and TypeSafe AI API key:
+
+```bash
+cp .env.example .env
+```
+
+### Step 4: Initialize the agent
+
+```bash
+uv run uipath init
+```
+
+### Step 5: Run the agent
+
+```bash
+uipath run main --input-file input.json
+```
+
+Try editing `input.json` to see both branches:
+
+* A calm, clearly-worded billing request (like the default input) -> the
+  agent auto-drafts a reply via the LLM Gateway and returns it directly.
+* An urgent or angry-sounding message (e.g. mentioning "urgent", "ASAP", or
+  "furious") -> the agent creates an Action Center task instead and returns
+  its task id.
+
+## How it works
+
+1. `triage_ticket` sends the ticket's subject/message to `Jev` (via
+   `TypeSafeClient.system_one`) as `state`, along with three typed
+   `questions` (`department: Choice`, `is_urgent: Noul`,
+   `frustration: Score`), and gets back a `TriageDecision`.
+2. `needs_escalation` checks the triage output against fixed thresholds
+   (urgency, frustration, routing confidence) in `main.py`.
+3. On escalation, `escalate_to_human` creates an Action Center QuickForm
+   task via `client.tasks.create_quickform(...)`, with the ticket and the
+   triage decision as task data for the reviewer.
+4. Otherwise, `draft_auto_reply` calls `client.llm.chat_completions(...)`
+   (UiPath LLM Gateway) with a department-specific system prompt to draft a
+   reply.
+
+## Evaluations
+
+`evaluations/eval-sets/default.json` exercises the Tier 1 routing/escalation
+logic against 11 tickets (billing, technical, and sales; calm and
+auto-replied vs. urgent/angry/low-confidence and escalated), using two
+evaluators:
+
+* `DepartmentRoutingEvaluator` (`evaluations/evaluators/department-routing.json`)
+  - a `uipath-multiclass-classification` evaluator checking
+  `triage.department` against the expected class.
+* `EscalationDecisionEvaluator` (`evaluations/evaluators/escalation-decision.json`)
+  - a `uipath-binary-classification` evaluator checking the `escalated`
+  boolean against the expected outcome.
+
+Two cases (`sales-demo-request-wrong-department`,
+`billing-refund-calm-wrong-escalation`) have deliberately wrong ground truth,
+mirroring the pattern in `classification_agent`, to demonstrate the
+evaluators catching a mismatch.
+
+Every case calls the real Jev API (`TYPESAFE_API_KEY`), and Tier 2 always
+runs too: `draft_auto_reply` and `escalate_to_human` call real UiPath
+services (LLM Gateway, Action Center), so the full eval set requires valid
+credentials in `.env` for all 11 cases, not just the ones the two
+evaluators grade.
+
+**Heads up:** none of the current evaluators grade `auto_reply` or
+`action_task_id`, but the ~3 cases whose triage escalates still create a
+real Action Center QuickForm task each run, and nothing in this sample
+cleans them up. Expect leftover tasks in Action Center after repeated
+`uipath eval` runs.
+
+```bash
+uipath eval
+```
+
+## Publish your coded agent
+
+Once tested locally, publish the agent to Orchestrator:
+
+```bash
+uipath pack
+uipath publish
+```

packages/uipath/samples/ticket-triage-agent/bindings.json

@@ -0,0 +1,4 @@
+{
+  "version": "2.0",
+  "resources": []
+}

packages/uipath/samples/ticket-triage-agent/evaluations/eval-sets/default.json

@@ -0,0 +1,143 @@
+{
+  "version": "1.0",
+  "id": "TicketTriageEval",
+  "name": "Ticket Triage Routing & Escalation Evaluation",
+  "evaluatorRefs": [
+    "DepartmentRoutingEvaluator",
+    "EscalationDecisionEvaluator"
+  ],
+  "evaluations": [
+    {
+      "id": "billing-refund-calm",
+      "name": "Billing - calm refund request (auto-reply)",
+      "inputs": {
+        "subject": "Duplicate charge on my account",
+        "message": "Hi, I was charged twice for order A-104. Could you please refund the duplicate charge? Thanks for your help."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "billing" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "billing-invoice-question",
+      "name": "Billing - invoice clarification (auto-reply)",
+      "inputs": {
+        "subject": "Question about my invoice",
+        "message": "I noticed my invoice this month includes a charge I do not recognize. Could you clarify the payment breakdown? Thanks."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "billing" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "technical-bug-report",
+      "name": "Technical - routine bug report (auto-reply)",
+      "inputs": {
+        "subject": "App crashes on login",
+        "message": "The app crashes every time I try to log in. This started after the last update. Please help me fix this bug."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "technical" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "technical-outage-urgent",
+      "name": "Technical - urgent outage (escalate: high urgency)",
+      "inputs": {
+        "subject": "URGENT: service down",
+        "message": "Our integration has been down for 3 days and this is unacceptable. We need this fixed immediately, it is critical."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "technical" },
+        "EscalationDecisionEvaluator": { "expectedClass": "true" }
+      }
+    },
+    {
+      "id": "sales-pricing-question",
+      "name": "Sales - pricing inquiry (auto-reply)",
+      "inputs": {
+        "subject": "Question about pricing plans",
+        "message": "Hi, I would like a quote for upgrading to your premium plan. Can you share pricing details?"
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "sales" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "sales-demo-request",
+      "name": "Sales - demo request (auto-reply)",
+      "inputs": {
+        "subject": "Requesting a demo",
+        "message": "We are considering your product and would like to schedule a demo to see the upgrade options."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "sales" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "billing-angry-customer",
+      "name": "Billing - angry customer (escalate: high frustration)",
+      "inputs": {
+        "subject": "This is ridiculous",
+        "message": "I have been charged the wrong amount again and I am furious, this is unacceptable and terrible service."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "billing" },
+        "EscalationDecisionEvaluator": { "expectedClass": "true" }
+      }
+    },
+    {
+      "id": "technical-calm-minor-bug",
+      "name": "Technical - minor, no-rush bug (auto-reply)",
+      "inputs": {
+        "subject": "Minor bug report",
+        "message": "Just a small bug: the export button is not working as expected. No rush, whenever you get a chance, thanks."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "technical" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "ambiguous-low-confidence",
+      "name": "Ambiguous ticket (escalate: low routing confidence)",
+      "inputs": {
+        "subject": "Quick question",
+        "message": "Can someone tell me more about how things work around here?"
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "billing" },
+        "EscalationDecisionEvaluator": { "expectedClass": "true" }
+      }
+    },
+    {
+      "id": "sales-demo-request-wrong-department",
+      "name": "Sales - demo request (DELIBERATELY WRONG: ground truth department set to 'technical')",
+      "inputs": {
+        "subject": "Requesting a demo",
+        "message": "We are considering your product and would like to schedule a demo to see the upgrade options."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "technical" },
+        "EscalationDecisionEvaluator": { "expectedClass": "false" }
+      }
+    },
+    {
+      "id": "billing-refund-calm-wrong-escalation",
+      "name": "Billing - calm refund request (DELIBERATELY WRONG: ground truth escalation set to 'true')",
+      "inputs": {
+        "subject": "Duplicate charge on my account",
+        "message": "Hi, I was charged twice for order A-104. Could you please refund the duplicate charge? Thanks for your help."
+      },
+      "evaluationCriterias": {
+        "DepartmentRoutingEvaluator": { "expectedClass": "billing" },
+        "EscalationDecisionEvaluator": { "expectedClass": "true" }
+      }
+    }
+  ]
+}

packages/uipath/samples/ticket-triage-agent/evaluations/evaluators/department-routing.json

@@ -0,0 +1,17 @@
+{
+  "version": "1.0",
+  "id": "DepartmentRoutingEvaluator",
+  "name": "DepartmentRoutingEvaluator",
+  "description": "Checks whether Tier 1 (Jev) routed the ticket to the expected department",
+  "evaluatorTypeId": "uipath-multiclass-classification",
+  "evaluatorConfig": {
+    "name": "DepartmentRoutingEvaluator",
+    "targetOutputKey": "triage.department",
+    "classes": ["billing", "technical", "sales"],
+    "metricType": "precision",
+    "averaging": "macro",
+    "defaultEvaluationCriteria": {
+      "expectedClass": "technical"
+    }
+  }
+}

packages/uipath/samples/ticket-triage-agent/evaluations/evaluators/escalation-decision.json

@@ -0,0 +1,16 @@
+{
+  "version": "1.0",
+  "id": "EscalationDecisionEvaluator",
+  "name": "EscalationDecisionEvaluator",
+  "description": "Checks whether the agent escalated to a human (Action Center) exactly when it should have",
+  "evaluatorTypeId": "uipath-binary-classification",
+  "evaluatorConfig": {
+    "name": "EscalationDecisionEvaluator",
+    "targetOutputKey": "escalated",
+    "positiveClass": "true",
+    "metricType": "f-score",
+    "defaultEvaluationCriteria": {
+      "expectedClass": "false"
+    }
+  }
+}

packages/uipath/samples/ticket-triage-agent/input.json

@@ -0,0 +1,4 @@
+{
+  "subject": "Duplicate charge on my account",
+  "message": "Hi, I was charged twice for order A-104. Could you please refund the duplicate charge? Thanks for your help."
+}