Agent Skills

Introduction

What is the Skills screen?

The Skills screen is where you control what your AI Assistant knows how to do and what external tools it can reach. It has three tabs along the top: Chat, Manage Agent Skills, and Manage MCPs. Together they let you teach the assistant specialised expertise, connect it to outside systems, and then talk to it directly.

Two words appear everywhere on this screen, so it helps to define them up front in plain language:

• A skill is a reusable set of written instructions and best practices that gets injected into the AI’s system prompt. Think of it as a cheat-sheet the AI reads before answering. For example, a Playwright skill tells the AI exactly how your team writes Playwright tests, what page-object patterns to use, and what to avoid. Each skill carries a few tags, a token cost, and an enabled/disabled switch.
• An MCP connector (Model Context Protocol) is a bridge to an external program or server that exposes real tools the AI can call, such as a browser automation server, a database, or a file-system server. Where a skill is knowledge, an MCP server is hands, it lets the AI actually do things outside Testver. Analogy: imagine hiring a brilliant new tester. A skill is the team handbook you hand them so they follow your conventions. An MCP connector is giving them keys and login badges to the systems they need (the test browser farm, the bug database, the file share). The Chat tab is the desk where you sit down and actually ask them to do the work, after deciding which handbooks and which keys they get for that conversation.

Note

In one sentence: the Skills screen lets you decide what the AI is good at (skills), what external systems it can touch (MCP connectors), and then put it to work (chat), all from one place.

Who this guide is for

This guide is written for everyday Testver users, no coding required to follow along. You will get the most out of it if you are:

• A QA engineer or test author who wants the AI to follow your framework’s conventions when it analyses code, writes scripts, or builds reports.
• A test lead or manager who wants quick reports, charts, and summaries generated through the Chat tab.
• An automation or DevOps engineer who wants to connect external tools (browser automation, databases, file systems) so the AI can act on real systems.
• Anyone curious about what the AI already knows and wanting to enable, disable, create, or import skills. You do not need to understand the Model Context Protocol internally. You only need to know how to fill in a couple of fields, which this guide walks through step by step.

Key terms

Term	What it means in Testver
Skill	A bundle of written instructions (a SKILL.md file: YAML front-matter plus a Markdown body) that is added to the AI’s prompt when enabled. Adds domain expertise.
Agent Skills	The collective name for all skills shown on the Manage Agent Skills tab.
Built-in skill	A skill that ships with Testver. Shown with a book icon and the Built-in source. Can be enabled/disabled and exported, but not deleted.
Custom skill	A skill you created or imported. Shown with a code icon and a purple custom badge. Can be edited, exported, and deleted.
Slug	The unique machine-readable id of a skill (e.g. playwright-best-practices). Shown in the skill’s detail view.
Token estimate	Roughly how much of the AI’s context budget a skill consumes when enabled (shown as ~N tok). More enabled skills means more tokens used.
MCP	Model Context Protocol, the standard Testver uses to connect external tool servers to the AI.
MCP server / connector	An external program Testver launches or connects to that exposes callable tools. Each shows a connection status and a tool count.
Transport	How Testver talks to an MCP server: STDIO (launches a local command) or SSE (connects to a URL over the network).
Preset / Catalog	A library of pre-configured MCP servers you can add with one click instead of typing settings by hand.
Built-in Tools	Categories of tools the AI already has inside Testver (e.g. file, browser, Testver data), independent of MCP. Toggled per chat conversation.
Session (chat)	One saved conversation in the Chat tab, with its own history and its own saved selection of skills, MCP servers, and tools.

How skills are used by the AI Assistant

Skills and connectors are not used automatically everywhere. They take effect through the Chat tab, where you pick which ones are active for the conversation you are about to have:

1. When a skill is enabled on the Manage Agent Skills tab, it becomes available to be picked in Chat (but is not auto-selected).
2. In the Chat tab’s Configure bar you pick exactly which enabled skills, which connected MCP servers, and which built-in tool categories the AI should use for this conversation.
3. When you send a message, Testver injects the selected skills’ instructions into the AI’s prompt and grants it access to the selected MCP tools and tool categories.
4. The AI then answers, calling tools as needed. You see each tool call appear live in the conversation timeline.

Note

Enabling a skill on the Manage tab is like unlocking it; selecting it in the Chat Configure bar is like equipping it for the current job. A skill that is disabled cannot be selected in chat at all.

Getting Started

Opening Skills

1. From the Testver navigation, open the Skills screen (route /skills).
2. It opens on the Chat tab by default.
3. Use the three tabs at the top, Chat, Manage Agent Skills, and Manage MCPs, to move between the areas.
4. Your active tab is remembered in the page address (the tab parameter), so you can bookmark or share a link straight to a particular tab.

The tabs

Tab	Icon	What you do here
Chat	Speech bubble	Talk to the AI agent. Pick which skills, MCP servers, and built-in tools it uses, upload files, and review saved conversations.
Manage Agent Skills	Gear	Browse, search, enable/disable, create, import, edit, export, and delete skills.
Manage MCPs	Plug	Add, configure, connect, test, enable/disable, and remove external MCP tool servers.

Note

When you are on the Chat tab, two extra controls appear at the top-right: a New Session button (starts a fresh conversation) and a History toggle (shows or hides the saved-sessions sidebar). These appear only on the Chat tab.

At a glance

Each management tab shows a summary header so you always know where you stand:

• On Manage Agent Skills: an N/M enabled counter, an estimate of total tokens active, and how many custom skills you have.
• On Manage MCPs: an N/M connected counter and the total number of tools available across all connected servers.
• On the Chat welcome screen: a one-line summary like X Skills · Y MCP · Z Tools Active reflecting your current selection.

The Skills List (Manage Agent Skills)

Overview

On the Manage Agent Skills tab, every skill appears as a card in a grid. The header explains the purpose: skills inject domain-specific best practices into the AI system prompt. A skill that is enabled is shown brightly with a cyan toggle; a disabled skill is dimmed.

Each card carries an icon indicating its source, a book icon for Built-in skills and a code icon for custom ones, plus the skill name, an optional version tag, the purple custom badge (custom skills only), a one-line description, up to three coloured tag chips, and the token estimate.

Anatomy of a skill card

Element	Where	What it does
Search box	Top bar	Filters cards by name, slug, description, tag, or framework as you type. A clear (x) button appears when text is present.
All Tags dropdown	Top bar	Narrows the list to skills carrying a chosen tag.
Source filter (All / Built-in / Custom)	Top bar	Shows all skills, only built-in skills, or only your custom skills.
Create button	Top bar	Opens the editor to write a brand-new custom skill from scratch.
Import button	Top bar	Opens the Import dialog to bring in a skill from a file, pasted text, or a GitHub URL.
Reset button	Top bar	Resets all skills to their default state and deletes every custom skill (confirmation required).
Enable/disable toggle	Each card (right)	Turns the skill on or off. Shows a spinner while saving.
Card body (click)	Each card	Opens the skill detail view (preview).
Export icon	Card hover	Downloads the skill as a .md file.
Edit icon	Card hover (custom only)	Opens the editor pre-filled with the skill’s content.
Delete icon	Card hover (custom only)	Permanently deletes the custom skill (confirmation required).
Token chip	Each card	Shows the skill’s approximate token cost (~N tok).

Tip

Tag chips are colour-coded by topic (for example testing, browser, e2e, python, java) so you can spot related skills at a glance. If a skill has more than three tags, the card shows a +N count.

Finding a skill

1. Find the skill you want using search, the tag dropdown, or the source filter.
2. Click its toggle on the right of the card. A right-pointing cyan toggle means enabled; a left-pointing grey toggle means disabled.
3. The card brightens when enabled and dims when disabled, and the header’s enabled counter and tokens active total update immediately. There is no separate bulk control surfaced on the cards, but the Reset action returns everything to defaults, and you select which enabled skills to actually use per conversation over in the Chat tab.

Note

Only enabled skills can be picked in the Chat Configure bar. If a skill you want is missing from Chat, check that it is enabled here first.

Viewing skill details

Click anywhere on a skill card (other than the toggle or hover icons) to open its detail view. This shows everything Testver knows about the skill:

• Name, description, and source (Built-in or Custom).
• Slug, optional Version and Author, and the Token Cost estimate.
• A Status button to enable or disable the skill right from the preview.
• Tags, Frameworks, and Recommended Tools chips, when present.
• The full Skill Instructions body, with a toggle between Rendered markdown and Raw source view, plus a character count.
• Action icons in the header to Export (all skills), and Edit / Delete (custom skills only).

Creating a skill

1. Click Create in the top bar to open the skill editor.
2. Write your skill in SKILL.md format: a YAML front-matter block (name, slug, description, and optionally version, author, tags, frameworks, tools) followed by a Markdown body of instructions.
3. Save. The new skill appears in the grid with a purple custom badge and can be enabled like any other.
4. Newly created skills are marked as your own, so you can edit or delete them later.

Tip

Keep custom skills focused. Because every enabled skill adds to the token cost shown in the header, several small, targeted skills are easier to mix and match than one giant one.

Importing a skill

Click Import to open the Import Skill dialog, which offers two modes via a toggle at the top:

Mode	How it works
File / Paste	Drag a .md file onto the drop zone, click to browse for one, or paste SKILL.md content directly into the text area. Only .md files are accepted; the dialog shows a green confirmation and character count once content is loaded.
GitHub URL	Paste a link to a skill file on GitHub. Supported forms include github.com/user/repo/blob/branch/path.md, raw.githubusercontent.com/…, and gist.github.com/…

1. Click Import to open the dialog.
2. Choose File / Paste or GitHub URL at the top.
3. Provide the content (drop/browse/paste a file, or enter a URL).
4. Click Import (or Import from URL). On success the dialog closes and the skill appears in your list; any error is shown in red at the bottom of the dialog.

Exporting a skill

• Export: hover a card and click the download icon (or use the Export icon in the detail view) to save the skill as a .md file you can share or back up. Available for every skill.
• Delete: hover a custom card and click the trash icon (or use Delete in the detail view). A confirmation dialog warns that the action is permanent. Built-in skills cannot be deleted, only disabled.

Caution

Reset does more than toggle switches: it returns all skills to their defaults and permanently deletes every custom skill you have created or imported. This cannot be undone, so export anything you want to keep first.

Using a skill

Skills are not ‘run’ on this tab, they are made available. To actually put a skill to work, switch to the Chat tab, open the Configure bar, select the enabled skill, and send a message. The AI then follows that skill’s instructions while answering.

MCP Connectors (Manage MCPs)

Overview

The Manage MCPs tab is where you connect external tool servers to the AI. Its header explains it: connect external MCP servers to extend your AI assistant with additional tools for browser automation, database access, and more. Each server you add appears as a card showing its name, transport type, connection status, and how many tools it exposes.

When the page has no servers yet, an empty state explains MCP and offers two buttons: Browse Catalog and Add Custom Server.

Anatomy of a connector

Element	Meaning
Server name + icon	The connector’s name, with a topic icon (e.g. browser, database, GitHub) where recognised.
Transport badge	STDIO (a local command Testver launches) or SSE (a URL Testver connects to).
Status badge	Connected (green), Connecting (amber spinner), Error (red), or Disconnected (grey).
Tool count	How many tools the server exposes once connected.
Command / URL line	For STDIO, the command and arguments; for SSE, the server URL.
Enable/disable toggle	Cyan toggle = enabled, grey = disabled. A disabled server cannot be connected and the card is dimmed.
Connect / Disconnect button	Opens or closes the live connection. Connect is unavailable while the server is disabled.
Test button	Probes the server and reports how many tools were discovered, without keeping the connection.
Edit / Delete (hover)	Edit the server’s configuration, or remove it entirely.
Tools expander	When connected, click to expand a list of each available tool’s name and description.

Note

The list refreshes its connection states automatically every few seconds while you are on the tab, so a status will update on its own shortly after a server connects or fails.

Adding a connector from the catalog

1. Click Add from Catalog (or Browse Catalog on the empty state).
2. Browse the MCP Server Catalog, organised into categories such as Browser & UI Testing, Database Testing, File System & Version Control, Development & Testing, and API & Contract Testing.
3. Each entry shows an icon, name, transport badge, description, its command, and a View docs link where available.
4. Click Add next to the one you want. Already-added entries show Added instead.
5. The server now appears as a card on the tab, ready to connect. Catalog entries cover popular tools including Playwright (Official and Community), Appium, OWASP ZAP, Trivy, Snyk, Lighthouse, k6 Load Testing, Axe and Pa11y accessibility, Percy and Applitools visual testing, DBHub, PostgreSQL, MySQL, MongoDB, SQLite, Filesystem, Git, GitHub, Slack, Jira, Linear, Docker, Kubernetes, GraphQL, OpenAPI/Swagger, Postman, and an ‘Everything’ demo server.

Adding a custom server

1. Click Add Server (or Add Custom Server on the empty state) to open the Add MCP Server dialog.
2. Give the server a Server Name.
3. Choose the Transport Type: STDIO for a local command, or SSE for a network URL.
4. Fill in the transport-specific fields (see the table below).
5. Optionally add Environment Variables (key/value pairs) the server needs, such as API tokens.
6. Optionally tick Auto-connect on server startup so Testver connects automatically next time it starts.
7. Click Add Server. (When editing an existing server the button reads Save Changes.)

Field	Applies to	What to enter
Server Name	All	A friendly label, e.g. Playwright MCP.
Transport Type	All	STDIO (local command) or SSE (network URL).
Command	STDIO	The executable to launch, e.g. npx.
Arguments (comma-separated)	STDIO	Command arguments, e.g. -y, @modelcontextprotocol/server-everything.
Server URL	SSE	The server endpoint, e.g. http://localhost:3001/sse.
Environment Variables	All	Optional KEY/value pairs; add as many rows as needed, remove with the x.
Auto-connect on server startup	All	Optional. When ticked, Testver connects this server automatically on launch.

Tip

If your arguments include --dsn (for database servers like DBHub), the dialog shows a helper panel with example connection-string formats for PostgreSQL, MySQL, MariaDB, SQL Server, and SQLite. Replace the sample value with your real connection string.

Note

The Add Server / Save Changes button stays disabled until the required fields are filled: a name is always required, plus a command for STDIO or a URL for SSE.

Managing connectors

Action	How	Result
Connect	Click Connect on the card	Opens a live connection; on success shows how many tools are available and the status turns green.
Disconnect	Click Disconnect on a connected card	Closes the connection; status returns to Disconnected.
Test	Click Test	Probes the server and reports the number of tools discovered (or the failure reason) without keeping the connection open.
View tools	Expand the tools list on a connected card	Lists each tool’s name and description.
Enable / Disable	Click the toggle	A disabled server is greyed out and cannot be connected.
Edit	Hover the card, click the pencil	Reopens the configuration dialog as Edit MCP Server.
Delete	Hover the card, click the trash	Asks for confirmation, then disconnects and removes the server’s configuration.

Caution

If a connection fails, the card shows a red Error status with the error message. Use Test to re-probe, check the command/URL and any required environment variables in Edit, then Connect again.

Skills Chat

Overview

The Chat tab is an AI agent you converse with directly. Unlike a plain chatbot, it can use the skills, MCP connectors, and built-in tools you select, so it can analyse your framework, generate documents and charts, query Testver data, and act through connected tools. When a conversation is empty it shows an AI Agent Ready welcome screen with quick actions and suggested prompts.

The Configure bar

Just above the message box sits a compact summary line, for example 🧩 N Skills · 🔌 N MCP · 🔧 All Tools. Click it (or the Configure button) to expand three pickers:

Picker	What you select	Notes
🧩 Skills	Which enabled skills the AI should follow for this conversation	Only skills enabled on the Manage Agent Skills tab appear. Click a chip to add or remove it.
🔌 MCP Servers	Which connected MCP servers to grant the AI	Only servers that are currently connected appear here. If none are connected it reads ‘No MCPs connected’.
🔧 Built-in Tools	Which categories of Testver’s own tools to allow	Each category shows its tool count. Use All or None shortcuts to include/exclude every category at once.

Note

You must have at least one source active, a skill, an MCP server, or at least one built-in tool category, before the AI will chat. Until then the message box is disabled and prompts you to Select skills, MCP, or tools to start.

Quick Actions

• Quick Actions on the welcome screen are one-click buttons (Analyze Framework, Generate Report, Create Chart, Find Bugs, Test Summary, Email Report, JIRA Tasks, Code Review) that send a ready-made prompt.
• Suggested Prompts are example questions you can click to get started.
• Prompt Templates (the clipboard icon by the message box) open a categorised flyout, Analysis, Reports, Charts, Management, Actions, each with several ready-to-edit prompts that drop into the message box.
• After the AI replies, follow-up suggestion chips appear so you can continue with a related request in one click.

Sending a message

1. Type your request in the message box. Press Enter to send, or Shift+Enter for a new line.
2. Attach files with the paperclip, or drag and drop them onto the input area. Many document, code, and image formats are accepted; uploaded files appear as removable chips.
3. If more than one AI provider is configured, use the model selector to choose which provider and model answers.
4. Watch the conversation timeline: the AI’s text streams in, and each tool it calls appears as a line you can click to inspect inputs and outputs. When the AI needs to do something sensitive, a Permission Required prompt appears in the timeline showing the tool name and arguments, with Allow and Deny buttons. Nothing runs until you choose. If a tool produces a downloadable document (Word, PDF, Excel, CSV), a download card appears in that tool’s detail view.

Starting a new session

• Click New Session (top-right) to start a fresh conversation.
• Click the History toggle to open the sidebar, which lists past sessions grouped as Pinned, Today, Yesterday, This Week, and Older, with a search box.
• Click a session to reopen it; the conversation and its saved skill/MCP/tool selections are restored.
• Use a session’s menu (three dots) to Pin to Top, Rename, or Delete it.
• If a restored session referenced skills or MCP servers that are no longer available, a warning is shown at the end of the conversation.

Tip

Each session remembers its own Configure-bar choices. Set up one session with your report-writing skills and another with your code-analysis skills, and switching between them restores the right setup automatically.

Common Tasks (How Do I…?)

I want to…	Do this
Turn a skill on or off	Manage Agent Skills tab → click the skill’s toggle.
See exactly what a skill tells the AI	Manage Agent Skills tab → click the card → read the Skill Instructions (Rendered or Raw).
Add my own skill	Manage Agent Skills tab → Create, write SKILL.md, save.
Bring in a skill from GitHub	Manage Agent Skills tab → Import → GitHub URL → paste link → Import from URL.
Back up a skill	Hover the card → download icon (Export as .md).
Remove a custom skill	Hover the card → trash icon → confirm.
Connect a popular external tool	Manage MCPs tab → Add from Catalog → Add next to the tool → Connect.
Add a tool that isn’t in the catalog	Manage MCPs tab → Add Server → fill in name/transport/command or URL → Add Server → Connect.
Check a connector works	Manage MCPs tab → Test on its card.
See what tools a connector provides	Connect it, then expand the tools list on its card.
Ask the AI to do something with my chosen tools	Chat tab → Configure → select skills/MCP/tools → type your request → Enter.
Get a report or chart from the AI	Chat tab → use a Quick Action or Prompt Template, e.g. Generate Report or Create Chart.
Reopen an old conversation	Chat tab → History toggle → pick the session.
Start fresh	Chat tab → New Session.

Tips & Best Practices

• Enable only the skills you actually use, the header’s tokens active total reminds you that every enabled skill adds to the AI’s context budget.
• Equip per job: keep skills enabled broadly, but in Chat select just the few that fit the task for cleaner, faster answers.
• Prefer the Catalog over manual setup for known tools, it fills in the command and arguments for you and links to docs.
• Always Test a new MCP server before relying on it, so you confirm it connects and exposes the tools you expect.
• Use Auto-connect for servers you depend on daily so they are ready the moment Testver starts.
• Store secrets in a connector’s Environment Variables rather than hard-coding them into arguments.
• Give chat sessions clear names and Pin the ones you return to, so your history stays easy to navigate.
• Read the Permission Required prompt before clicking Allow, it shows exactly which tool will run and with what arguments.
• Export custom skills you care about before using Reset, which deletes them permanently.

Troubleshooting & FAQ

Question / Problem	Answer
A skill I enabled doesn’t appear in the Chat Configure bar.	Confirm it is still enabled on Manage Agent Skills. Only enabled skills are selectable in Chat.
The chat message box is greyed out.	You have nothing selected. Open Configure and select at least one skill, one connected MCP server, or one built-in tool category.
My MCP server isn’t listed in the Chat MCP picker.	The Chat picker only lists connected servers. Go to Manage MCPs and click Connect first.
A connector shows a red Error status.	Read the error on the card, then Edit to check the command/URL and environment variables, Test to re-probe, and Connect again.
I can’t delete a built-in skill.	Built-in skills can only be disabled, not deleted. Only custom skills have a delete option.
Import failed.	For files, ensure it is a .md in SKILL.md format (YAML front-matter + Markdown). For URLs, use a supported GitHub/raw/gist link. The error reason is shown at the bottom of the dialog.
The AI paused and asked for permission.	That is by design for sensitive tools. Review the tool name and arguments, then click Allow or Deny.
Where did my generated document go?	Open the relevant tool line in the conversation timeline; a download card appears there for Word, PDF, Excel, and CSV outputs.
My old session lost some skills/MCP servers.	If those skills or servers were removed since the session was saved, a warning appears at the end of the conversation noting they are no longer available.
What’s the difference between a skill and an MCP connector?	A skill is knowledge injected into the AI’s prompt; an MCP connector is access to an external tool the AI can call. You can use either or both together.
Will enabling lots of skills slow things down?	It increases the token cost (shown in the header) and the context the AI must read. Enable broadly but select narrowly per conversation for best results.

Related

• AI Assistant — uses skills + MCPs transparently.
• AI Test Gen — uses test-generation skills.
• Settings → LLM — configure providers skills run on.