Skip to content
Last updated

Skills & Marketplace

WebDesktopMobile

Skills are what separate a generic AI conversation from one that knows your Treasure AI environment. A skill teaches the AI how to do something specific — build a CDP journey in YAML, configure a Treasure Engage campaign, or manage a Treasure Workflow — so its output lands much closer to expert work without you writing the prompt yourself.

Note

Skills are instruction packages, not code plugins. They don't run on your machine or access your data directly. Instead, they give the AI a structured playbook — rules, examples, and tool-usage patterns — that it follows when you ask a question in that domain. All data access still goes through the same secure tool calls (queries, API requests) visible in the chat stream.

Objective

Understand what skills are, how they make AI responses more accurate, and how to manage them through plugins across Web, Desktop, and Mobile.

Prerequisites


Why Skills Matter

Without skills, the AI relies on its general training. It can write SQL, but it may not know about td_interval() or partition pruning. It can draft a segment definition, but it's unlikely to follow the exact YAML schema your CDP expects. Skills close that gap.

Without SkillsWith Skills Active
Generic SQL that may not use TD functionsTrino SQL with td_interval(), td_time_range(), and partition pruning
Approximate segment structureCDP segment YAML in the required schema, with the correct operators and field names
General workflow advice.dig files that use td> operators, _retry, and _error handling
Basic data explorationQueries scoped to your default database with session context

For Marketers and business users: Skills mean you can ask "build a segment for users who purchased in the last 30 days" and the AI drafts a CDP segment definition in the required YAML schema — a concrete starting point, not a generic template. Check that it captures what you meant, then push.

For Data Analysts and engineers: Skills mean the AI writes queries using the functions and patterns specific to Treasure AI's Trino and Hive engines, so they need far less hand-editing before they run.

Skills improve results — they don't guarantee them

Skills raise the quality and consistency of what the AI produces, but its responses are probabilistic, not deterministic — the same request can yield different output, and a result can look right without matching what you meant. Treat generated queries, segment and journey definitions, and workflows as drafts: review them, and validate before you run or push.

How Skills Work

A skill is a folder with a SKILL.md file of structured instructions. The AI is aware of every available skill from its name and description, and reads a skill's full instructions only when your request calls for it. Once a skill is in play, the rest of that chat's responses follow its guidance.

Skills reach the AI through plugins: a plugin bundles one or more skills, and plugins come from marketplaces (see Plugin Marketplace). You don't pick individual skills — you enable the plugins you want, and the AI applies their skills on its own. The lifecycle is:

  1. Enablement — A plugin is enabled for you, either by default or in Settings → Marketplace. Its skills become available.
  2. Awareness — The AI knows each available skill by its name and description — a lightweight index — so it can tell when one is relevant. It doesn't hold every skill's full text.
  3. Application — When your request matches a skill, the AI reads that skill's full instructions and follows its rules, examples, and tool patterns — no manual selection.
  4. Response — The AI answers using the skill's guidance.
Technical Note

Which plugins are active is a separate matter from how a skill loads: Studio resolves the set of enabled plugins when a chat starts, so enabling or disabling a plugin takes effect in your next chat — not in a conversation that is already open.


Pre-Installed Skills

Studio ships with the td-skills marketplace pre-installed and its tdx-skills plugin enabled by default. This plugin delivers a curated set of Treasure AI skills for CDP, AI agents, workflows, campaigns, and the tdx CLI, maintained by the platform team.

CategorySkillsWhat They Do
CDP & Segmentsparent-segment, segment, validate-segment, parent-segment-analysisBuild, validate, and analyze CDP parent and child segments
Journeys & Activationsjourney, validate-journey, connector-configDesign customer journeys and configure activation connectors
CampaignsengageBuild and deploy Treasure Engage email and push campaigns
AI Agentsfoundry-agent, foundry-agent-prompt, foundry-agent-testBuild, prompt, and test AI Agent Foundry agents
Workflows & CLIworkflow, tdx-basicManage Treasure Workflows and run core tdx CLI operations

For the full catalog of every tdx-skills skill with descriptions and persona mapping, see Supported Skills Catalog.

Other plugins in the td-skills marketplace — SQL query skills, workflow authoring, real-time personalization, and more — are available too, but are not enabled by default. Enable them in Settings → Marketplace (see Plugin Marketplace).

Free accounts

Free accounts ship with a different default: the td-plg-skills marketplace and its treasure-ai-experience plugin, in place of tdx-skills.


How Skills Are Applied

Once a plugin is enabled, the AI applies its skills automatically whenever your request falls within a skill's domain — you don't select or invoke anything, you just describe what you want. There is no per-chat skill picker; skills are matched to your request as you work.

Example:

With the segment skill available (part of the default tdx-skills plugin):

You: Build a segment of users who purchased in the last 30 days

The AI applies the segment skill's rules to draft a CDP segment definition — following the schema's operators and field names — without you having to mention the skill. Review it against what you intended before you push.

You can also invoke any installed plugin's skill explicitly with a slash command — /{plugin-name}:<skill>, for example /tdx-skills:segment. This works for the pre-installed plugins and for skills your teammates share (see Sharing Custom Skills).

On Mobile (iOS), plugin management isn't available in-app, but every plugin you enabled on Web or Desktop applies to your Mobile chats. Describe what you want in natural language and the AI applies the relevant skills.


Managing Which Skills Are Active

You don't choose individual skills — you control which plugins are enabled, and the AI applies their skills automatically. Plugin management lives in Settings → Marketplace on Web and Desktop.

  • Enable or disable a plugin to add or remove its skills from your chats. The default tdx-skills plugin is enabled out of the box.
  • Register more marketplaces to install additional plugins, if your account policy allows it.
  • Changes take effect the next time you start a chat.

For the full workflow — registering marketplaces, enabling plugins, and the administrator controls that govern access — see Plugin Marketplace. To build your own skills and share them privately with your account without a public repository, see Sharing Custom Skills.

Web and Desktop only

The marketplace and plugin management interface is available on Web and Desktop. Plugins you enable there apply to your chats on all platforms, including Mobile.


How Skills Shape AI Behavior

When a skill is active, it shapes how the AI works in a few consistent ways:

Domain Accuracy

Active skills feed Treasure AI-specific knowledge into the AI's reasoning, steering it toward the right conventions. For example, with the trino skill active (from the sql-skills plugin, which you enable in the marketplace), the AI is far more likely to:

  • use td_interval() for time filtering instead of a generic WHERE clause
  • add partition-pruning hints that keep queries fast
  • reach for TD-specific functions like td_time_range() and td_sessionize() where they fit

Output Validation

Skills that generate configuration (segments, journeys, workflows) carry validation rules that guide the AI to:

  • follow the YAML schema the CDP API expects
  • use the right operator types (Equal, In, Between, TimeWithinPast)
  • check field names and time units before presenting the output

Tool Orchestration

Skills guide the AI in choosing the right sequence of tool calls. For example, the workflow skill teaches the AI to:

  1. Check tdx wf sessions for recent runs before debugging
  2. Use tdx wf attempt logs to inspect failures
  3. Suggest _retry blocks for transient errors

Verification

After reading this guide, you can:

  • Explain what a skill is and how it improves AI responses
  • Explain how skills reach the AI through plugins and marketplaces
  • Understand how the AI automatically applies the skills of enabled plugins
  • Enable or disable a plugin in Settings → Marketplace (see Plugin Marketplace)
  • Name the skills the default tdx-skills plugin delivers (see Supported Skills Catalog)

Troubleshooting

IssueSolution
Plugin changes not taking effectStart a new chat — enabled plugins are resolved when a chat starts, so an open conversation keeps the set it began with
A skill isn't being appliedConfirm its plugin is enabled in Settings → Marketplace, then make your request more specific so it clearly matches the skill's domain
No skills seem availableOpen Settings → Marketplace and verify at least one plugin is enabled — the default tdx-skills plugin should be on
Can't manage plugins on MobilePlugin management is Web and Desktop only. Enable plugins there; they apply to your Mobile chats automatically

Next Steps