AgentRun.Umbraco 1.2.0

AgentRun.Umbraco

AI-powered workflow engine for Umbraco CMS -- define multi-step agent workflows in YAML.

AgentRun ships with three example workflows, including an Umbraco Content Audit that analyses the content in your Umbraco instance directly from the backoffice -- no external URLs needed. After installing the package and configuring an Umbraco.AI profile, open Agent Workflows, click the Content Audit, and click Start. The agent reads your published content tree and produces a prioritised audit report.

Also included: Content Quality Audit and Accessibility Quick-Scan -- paste one or more public URLs and the agent fetches, scores, and reports on content quality or WCAG 2.1 AA accessibility issues.

What's Inside

AgentRun is an engine, not just a set of workflows. The shipped examples are a starting point; building your own is the intended path. Headline capabilities:

• Extensible tool surface. 10 shipped tools covering HTTP fetch, file I/O, published Umbraco content (list_content, get_content, list_content_types, search_content via Examine), an Umbraco.AI Context reader (get_ai_context), and web search (web_search, Brave + Tavily). Custom tools plug in via a simple .NET interface -- register once through Umbraco's composer system and every workflow can declare them.
• Prompt variables. Agent prompts can include {today}, {instance_id}, {step_index}, {previous_artifacts}, and workflow-declared config: tokens -- substituted deterministically at prompt-assembly time, so dates, IDs, and config values are never hallucinated.
• Provider-aware prompt caching. Cache-friendly prompts get ~90% discounts on Anthropic and automatic caching on OpenAI / Azure OpenAI, without YAML changes -- just a stable- first authoring discipline documented in the guide.
• Per-step model profiles. Route each step to the Umbraco.AI profile that fits -- Sonnet-class for high-volume scanning, Opus-class for the visible report -- and keep the cost / quality balance tunable per workflow.
• Agent Sanctum pattern. For agents whose identity matters (archivist voice, evidence- first analyst), split PERSONA / CREED / CAPABILITIES into ship-time sidecars. The engine assembles them into a cache-stable preamble above the per-turn runtime context.
• Umbraco Content Audit v2. The shipped audit now extracts real body prose (Block List, Block Grid, Rich Text) via a 3-tier property priority chain and scores an optional Brand pillar when an Umbraco.AI Context is configured -- six-pillar audit by default, seven with a brand voice attached.

See the Workflow Authoring Guide for the full tool reference and walkthroughs.

Quick Start

Prerequisites

• Umbraco 17+
• .NET 10
• At least one Umbraco.AI provider package (Umbraco.AI.Anthropic or Umbraco.AI.OpenAI) with a configured profile

1. Set up Umbraco first

If you're adding AgentRun to a new site, run the Umbraco install wizard first (dotnet run, complete setup in the browser, then stop the site). Umbraco.AI runs database migrations on startup and needs an existing database.

2. Install the packages

dotnet add package Umbraco.AI
dotnet add package Umbraco.AI.Anthropic
dotnet add package AgentRun.Umbraco --version 1.2.0

All three packages auto-register via Umbraco's composer system -- no Program.cs changes needed. Installing Umbraco.AI as a direct dependency ensures it survives if you later remove AgentRun.

3. Run the site

dotnet run

On first startup, AgentRun automatically:

• Copies three example workflows to App_Data/AgentRun.Umbraco/workflows/
• Grants the AgentRun section permission to the Administrators user group

No manual file copying or permission setup needed. Log out and back in to pick up the new section permission.

The three example workflows:

4. Configure Umbraco.AI

In the backoffice, go to the AI section:

1. Create a Connection -- select your provider (e.g., Anthropic), enter your API key
2. Create a Chat Profile (Profiles > Create > Chat) -- select the connection, choose a model, set an alias

AgentRun auto-detects your Umbraco.AI profile -- no YAML editing needed for first run. If you have multiple profiles, you can set default_profile in each workflow's workflow.yaml for deterministic selection.

5. Open Agent Workflows

Navigate to the Agent Workflows section in the backoffice. All three example workflows should appear. Start with the Umbraco Content Audit -- it reads your site's own content, so there's nothing to configure beyond the profile. For the URL-based workflows, click Start and paste one or more URLs when prompted.

Trying the Content Audit's Brand Pillar (Optional)

The Umbraco Content Audit runs as a 6-pillar audit by default. To enable the optional 7th Brand pillar -- which scores content alignment against your site's brand voice:

1. In the backoffice AI > Contexts section, create a Context with at least one brand-voice resource (populate toneDescription, targetAudience, and styleGuidelines)
2. Open App_Data/AgentRun.Umbraco/workflows/umbraco-content-audit/workflow.yaml
3. Set config.brand_voice_context: to the alias of the Context you created (leaving it as "" keeps the 6-pillar audit)
4. Restart the site so the workflow manifest refreshes
5. Run the audit -- the scanner validates the alias before starting, calls get_ai_context to load the brand voice, and scores a 7th Brand pillar against the prose extracted from each node's body-copy property

If the alias doesn't resolve, the scanner halts with a clear configuration error rather than silently falling back. See the Workflow Authoring Guide for the full pattern, including how to use it in your own workflows.

Creating Your Own Workflows

Workflows are YAML definitions paired with agent prompt markdown files. See the Workflow Authoring Guide for the full field reference, tool documentation, and a step-by-step walkthrough of creating a custom workflow.

The shipped example workflows in App_Data/AgentRun.Umbraco/workflows/ are also useful references -- read them alongside the guide.

Configuration

AgentRun works out of the box with sensible defaults. Most users only need an Umbraco.AI profile configured. For advanced use cases, add an AgentRun section to appsettings.json:

{
  "AgentRun": {
    "DefaultProfile": "your-profile-alias",
    "DataRootPath": "App_Data/AgentRun.Umbraco/instances/",
    "WorkflowPath": "App_Data/AgentRun.Umbraco/workflows/",
    "ToolDefaults": {
      "FetchUrl": {
        "MaxResponseBytes": 1048576,
        "TimeoutSeconds": 15
      },
      "ReadFile": {
        "MaxResponseBytes": 262144
      },
      "ToolLoop": {
        "UserMessageTimeoutSeconds": 300
      }
    },
    "ToolLimits": {
      "FetchUrl": {
        "MaxResponseBytesCeiling": 5242880,
        "TimeoutSecondsCeiling": 120
      }
    }
  }
}

All paths are relative to your site's content root. DefaultProfile sets the fallback LLM profile; individual workflows and steps can override it.

ToolDefaults sets site-wide default tuning values. ToolLimits sets hard ceilings that workflows cannot exceed -- any workflow declaring values above the ceiling is rejected at load time. See the Workflow Authoring Guide for the full resolution chain.

Known Limitations

• Single concurrent instance per workflow
• Disk-based persistence only (YAML + JSONL) -- no database
• Interactive mode is the primary UX; autonomous mode is functional but secondary
• Anthropic cache_control marker translation through Umbraco.AI is a planned follow-up; caching still works automatically on OpenAI / Azure OpenAI once prompts cross the provider size threshold
• Provider compatibility: Anthropic is the most thoroughly exercised provider to date. OpenAI and Azure OpenAI are supported through Umbraco.AI and work in practice; please report any provider-specific error scenarios you hit

Uninstalling

AgentRun creates no database tables, writes no configuration entries, and stores all runtime data under App_Data/AgentRun.Umbraco/. To remove it cleanly:

dotnet remove package AgentRun.Umbraco

Then optionally delete the data folder:

rm -rf App_Data/AgentRun.Umbraco/

No other cleanup is needed.

Licence

AgentRun.Umbraco is licensed under the Apache License 2.0 (SPDX: Apache-2.0).

Copyright 2026 Adam Shallcross.

Third-party components included or referenced by this package are acknowledged in NOTICE, all used under permissive licences compatible with Apache 2.0.