ThE Idea is that HighLevel should expand the Skills Builder inside Managed Agents so that a Skill is not limited to just written instructions/prompts. A Skill should be able to exist as a complete skill package containing: Skill = Instructions + References + Examples + Assets + Templates + Executable Scripts This is similar to how modern AI skill systems such as Claude Skills can package supporting resources alongside the primary skill instructions. The Problem Right now, when we think about creating an AI Skill, it is easy to treat the Skill as essentially a sophisticated prompt: > “When asked to do X, follow these instructions.” But many real business processes cannot be represented properly by instructions alone. For example, imagine creating a Meta Ads Media Buying Skill . The instructions might explain: How to create a campaign How to select an objective How to structure ad sets How to choose a bidding strategy How to optimize campaigns How to troubleshoot problems But the agent may also need: Screenshots showing what good campaigns look like Reference documentation SOPs Example campaigns Example ad creative JSON examples API documentation Brand guidelines Templates Checklists Code/scripts that can be executed Expected output examples These shouldn't all have to be stuffed into one enormous prompt. ## Proposed Solution: Skill Resources Inside the Managed Agents Skills Builder, allow every Skill to have Resources attached to it. For example: ```text Meta Ads Media Buying │ ├── Skill Instructions │ ├── References │ ├── Meta Ads SOP.pdf │ ├── Optimization Rules.md │ └── Troubleshooting Guide.md │ ├── Examples │ ├── Winning Campaign Example.json │ ├── Campaign Structure Example.md │ └── Example Report.pdf │ ├── Images │ ├── Campaign Structure.png │ ├── Ads Manager Example.png │ └── Creative Examples/ │ ├── Templates │ ├── Campaign Brief │ ├── Optimization Report │ └── Client Report │ └── Scripts ├── analyze-performance.py └── calculate-budget.js The Skill instructions could reference these resources directly. For example: > Before optimizing a campaign, read `References/Optimization Rules.md`. Or: > Structure the output using `Templates/Client Report`. Or: > Compare the creative against the examples inside `Images/Creative Examples/`. Or: > Run `Scripts/analyze-performance.py` against the campaign data before making your recommendation. ## Resource Types The Skills Builder could support several resource types: - **Reference Files** — PDFs, Markdown, TXT, DOCX, CSV, JSON, etc. - **Images** — screenshots, diagrams, examples, visual references - **Examples** — examples of good inputs and outputs - **Templates** — reusable structures the agent should follow - **Code/Scripts** — Python, JavaScript, etc. that the agent can execute when the Skill requires deterministic processing - **Schemas** — JSON schemas, API structures, expected data formats - **Knowledge/Documentation** — SOPs, policies, API documentation, internal documentation Resource Picker Inside the Skill Editor When writing Skill instructions, there should also be an easy way to reference a resource. For example, typing: `@` could bring up: ```text @Meta Ads SOP @Optimization Rules @Campaign Structure Example @Client Report Template @analyze-performance.py This would work similarly to how we already prompt AI systems using attached references and context. The difference is that these resources would live permanently inside the Skill instead of having to be attached every time the agent is prompted. Why This Matters This would fundamentally change Managed Agents from being primarily prompt-based agents into agents capable of carrying reusable organizational expertise. Businesses already operate using: SOPs + examples + templates + documentation + tools + scripts. AI Skills should be able to work the same way. Instead of trying to describe an entire business process inside one massive prompt, businesses could package everything necessary to perform that process into a reusable Skill. For example: ```text Media Buying ├── Meta Ads ├── Google Ads ├── TikTok Ads └── LinkedIn Ads Each Skill could contain its own instructions, documentation, screenshots, examples, templates and executable resources. Then those Skills could have more specialized Skills underneath them: ```text Meta Ads ├── Account Setup ├── Campaign Creation ├── Budget & Bidding ├── Optimization └── Troubleshooting ├── Account Restrictions ├── Tracking Problems ├── Poor Performance └── Ad Rejections This makes it possible to turn an organization's SOP library into an AI Skill library . Bigger Opportunity Long term, this could become one of the most powerful parts of Managed Agents. A business could effectively teach an agent how the company operates by giving it reusable Skills containing: Instructions + Knowledge + Examples + Assets + Templates + Code + Tools Instead of: > “Here is a prompt telling the AI what to do.” Managed Agents becomes: > “Here is everything the AI needs to know, reference, and use to perform this job correctly.” That would make Skills much closer to deployable AI capabilities rather than simply saved prompts.
