How skills and agents work in AWS Quick

How Skills and Agents Work in AWS Quick

AWS Quick is an AI‑powered assistant that lets you automate everyday work—status reports, ticket creation, data analysis—by speaking or typing natural language. Under the hood, Quick orchestrates agents, skills, and data connectors to understand your request, fetch the right information, and take action. This guide walks through those core concepts and shows how to build reusable skills that your whole team can share.

---

1. Agents: The Conversational Engine

An agent is a configurable AI persona that drives a conversation. It has three key parts:

1. Instructions – a prompt that defines the agent’s purpose and tone.
2. Knowledge sources – one or more spaces (collections of documents, dashboards, or knowledge bases) the agent can draw from.
3. Connectors – integrations that let the agent read from or write to external tools such as Slack, Jira, Salesforce, or a local MCP server.

Built‑in vs. custom agents

• Quick (built‑in): Every user gets a default agent that can access any resource you have permission to use. It’s ideal for ad‑hoc queries because you don’t need to pre‑configure a scope.
• Custom agents: You create a dedicated agent for a specific function—e.g., an HR onboarding assistant that only sees the HR space and the employee‑directory connector. After testing, you publish the agent so teammates can invoke it directly.

Agents run in a managed loop that follows the ReAct (reason‑act) pattern: they reason about the goal, select the appropriate skill or connector, act, and then iterate until they produce a final answer.

---

2. Data Foundations: Spaces, Knowledge Bases, Datasets, and Connectors

Spaces

A space is a curated container for the documents, dashboards, and other assets an agent may need. You can create personal spaces, team spaces, or organization‑wide spaces. Adding a space to an agent makes its contents automatically available during a conversation.

Knowledge bases

These are indexed collections of content (Google Drive, SharePoint, S3, etc.) optimized for fast retrieval. Quick syncs them on a schedule, keeping the information fresh for question‑answering.

Datasets & dashboards (QuickSight)

Structured data lives in datasets that Quick can query with natural language. A dashboard is the visual representation of a dataset analysis, which agents can embed in responses or update on demand.

Connectors

Connectors bridge Quick to external services. Cloud connectors (Outlook, Slack, Jira, Salesforce) work wherever you run Quick, while local connectors run on a machine you control, exposing private APIs or self‑hosted MCP servers.

Together, these data layers let agents ground their answers in the exact information you need, rather than relying on generic web search results.

---

3. Skills: Reusable Action Packages

A skill is a self‑contained set of instructions that teaches an agent how to perform a specific task. Skills follow the open Agent Skills specification (originating from Anthropic) and consist of:

• A `SKILL.md` file with front‑matter metadata (name, description, inputs, outputs).
• Optional scripts (Python, JavaScript, Bash) that implement the logic.
• Any supporting assets (templates, sample data).

System vs. custom skills

• System skills are built into Quick (e.g., generating a Word document, browsing the web, transcribing audio). You can enable or disable them, but you cannot edit or share them.
• Custom skills are authored by you or your team. Once created, you can publish them to the Quick catalog, making them discoverable and reusable across the organization.

Typical use cases

| Use case | Example skill | |----------|---------------| | Document generation | Create a weekly status report from a Markdown template and data pulled from a QuickSight dataset. | | Cross‑tool workflow | Pull the latest Jira tickets, summarize them, and post the summary to a Slack channel. | | Data analysis | Run a Python script that calculates churn rate, then render a chart with JavaScript and embed it in a QuickSight dashboard. |

Because skills are versioned folders, you can iterate on them, test in a sandbox agent, and roll out updates without breaking existing workflows.

---

4. Putting It All Together: A Quick Build Flow

1. Define the problem – Identify the repeatable task (e.g., “generate a quarterly sales deck”).
2. Create a space – Gather the source documents, datasets, and any templates needed.
3. Add connectors – Link to the tools that will receive the output (e.g., PowerPoint, email).
4. Author a skill – Write `SKILL.md` with clear input parameters (quarter, region) and embed the script that assembles the deck.
5. Configure an agent – Attach the space and relevant connectors, then assign the new skill.
6. Test & publish – Run a few prompts, refine the skill, then publish the agent for team use.

Once published, teammates can invoke the agent by name (e.g., `@SalesDeckBot generate Q3 deck for APAC`) and receive a fully‑formed deliverable without manual copy‑pasting.

---

Closing Thoughts

AWS Quick turns repetitive, knowledge‑heavy work into a conversational experience by tightly coupling agents, skills, and connected data. By structuring your data in spaces, building reusable skills, and configuring agents with the right permissions, you can scale automation across the organization while keeping control and security. Start with a simple skill—like summarizing daily Jira updates—and expand to more complex pipelines as you become comfortable with the ReAct loop. The result is a unified AI assistant that works for your team, not just with it.