Browse documentation

Skills

Package reusable guidance an agent can load when its task calls for it.

A skill is a named, reusable set of instructions an agent loads when its task calls for it, so you do not have to repeat the same guidance in every agent's instructions. Where a tool lets an agent act, a skill shapes how it handles a particular kind of task.

SPARK ships a catalogue of NGT Skills, written by NGT and covering investment operations procedures, controls, regulation, reporting and how to build in SPARK itself. Your workspace can add its own custom skills, written by people or by agents. What the library covers is in The NGT Skills catalogue, and how to choose skills for an agent in Give an agent NGT Skills.

Open the Skills page

Open Skills in Agent Studio: on the Settings tab it sits under Agent Studio, and on the Ecosystem tab you rest the pointer on Studio and pick it. You can also open it from an agent's Skills section with Manage skills.

The page shows every skill an agent can load, as cards. Your workspace's own skills come first, under Custom skills. NGT Skills follow on shelves by label family: Category, Process, Role, Jurisdiction, Asset type and Kind, each label a card with how many skills it holds. The categories are Analytics, Controls, Data, Fund Industry, Mathematics, Oversight, Private Assets, Regulation, Reporting and Spark Ecosystem, each in its own colour, the same colours the Tools page uses. A custom skill counts in its category too. Use the search (Search skills) to filter.

SkillsReusable instructions an agent loads when its task calls for them.NewCustom skillsMonth-end NAV checksThe checks a NAV packpasses before it goes t…CategoryAnalytics26 skillsControls24 skillsData35 skillsFund Industry62 skillsMathematics43 skillsOversight53 skillsPrivate Assets110 skillsRegulation57 skillsReporting38 skillsSpark Ecosystem54 skills
The Skills page: your custom skills first, then NGT Skills by label.

Click a card to open its label. The family's labels run across the top as tabs, each with its count; the label's skills are listed on the left, with Search skills and toggles to narrow them by kind and asset type; and the skill you pick is read in full on the right. The reading pane shows the skill's name, its category and kind, what it is for, its Role, Process, Jurisdiction and Asset type, its Version and Id, and its Instructions, with a Pretty view and a Raw view of the underlying Markdown.

Create a custom skill

  1. On the Skills page, click New.
  2. Fill in the New skill dialog (Instructions an agent loads when its task calls for them.):
FieldPurpose
Skill nameThe name shown in the catalogue. Required, and unique in the workspace.
DescriptionA one-line summary shown in the catalogue.
DomainAn optional label used to group the skill.
InstructionsThe Markdown the agent reads when it loads the skill. Required.
  1. Click Create. The new skill opens among your custom skills.

Duplicate and modify an NGT skill

NGT Skills cannot be edited, but you can copy one and change the copy. Open the skill and click Duplicate and modify. The confirmation, Duplicate and modify this skill?, says what happens: A copy is added to your workspace's own skills and opens for you to edit. The original stays as it is. Click Duplicate.

CategoryProcessRoleJurisdictionAsset typeKindAnalytics26Controls24Data35Fund Industry62Mathematics43Oversight53Private Assets110Regulation57Reporting38Spark Ecosystem54AnalyticsSearch skillsProcedure18Reference8Liquid funds15Private assets7Procedure18Annual Risk Limit ReviewUse to run the annual review of a fund's risk limits:limit history and utilisation, realised risk against the...Compliance, RiskAttribution Tie-OutAUM Roll-Forward and Net FlowsIndependent NAV EstimateDuplicate and modifyAnnual Risk Limit ReviewAnalyticsProcedureUse to run the annual review of a fund's risk limits: limithistory and utilisation, realised risk against the disclosedrisk profile, coverage of market, credit, liquidity,counterparty and operational risk, new levels back-testedon last year's positions, approval, loading and anyprospectus or KID change. Words: tracking error limit, VaRlimit, warning level. The daily limit check:ngt.analytics.compliance_checks.RolePortfolio manager, Risk managerProcessCompliance, RiskJurisdictionEUAsset typeLiquid fundsVersion1.0.0Idngt.analytics.risk_limit_reviewInstructionsWhat it is
An NGT skill read in full, with Duplicate and modify to make it your own.

Edit, duplicate or delete a custom skill

Open one of your custom skills. Its reading pane offers:

  • Duplicate: makes another custom copy.
  • Edit: reopens the dialog, titled Edit skill, then Save.
  • The trash icon (Delete skill): removes the skill for good. The confirmation warns that it is also unattached from any agents using it.

Let agents write skills

Agents can create and update your workspace's custom skills with their own tools, for example when you ask Spark to turn a procedure you described into a skill. An agent can change only custom skills, never NGT Skills, and it cannot delete a skill: deleting stays a person's decision on the Skills page. If you ask for a name the workspace already uses, the agent updates that skill or picks another name.

The tools are Spark Create Skill (Creates one of the workspace's own skills, or updates one, for agents to follow.) and Spark Read Skill, listed under Agents on the Tools page.

Attach a skill to an agent

  1. In Agent Studio, open an agent.
  2. In the Skills section, click + (Add skills), or Add while it is empty.
  3. The Select skills dialog (Tick the skills this agent can use.) shows the same shelves and reading pane as the Skills page, with a tick on each skill. Tick the skills this agent can use, then click Apply.

Attached skills appear as chips in the Skills section; use the × on a chip to remove one. Custom and NGT Skills sit in the same list and attach the same way.

Reach for a skill when the same guidance should apply across several agents or threads. Keep one-off instructions in the agent's own Instructions section instead.

Skills that come with tools

Spark's tools are linked to the skills that say how to use them. When you give an agent a tool, it also holds that tool's main skill, so it can read it before it writes, without you attaching it. The agent's own skills always come first in its catalogue.

When an agent wants a skill it was not given

Skills often point to each other. An agent reads its own skills freely, but when it asks for a skill it was not given, the thread pauses on a card, Skills the agent wasn't given, so you choose whether it may. Approving lets it read that skill, and any other, until the turn ends; rejecting tells it to stay with its own skills. Where nobody can answer, such as a workflow's Ask Agent step, an agent reads only its own skills.

How skills show up in a thread

When an agent loads a skill, the answer's steps show it as a line with the skill's name and how long it took. Click the line to open the skill as the agent read it (The skill the agent loaded, as it read it.). In Monitoring, the Tools & skills section counts how often each skill is loaded.

CMClaire Martin2 minutes agoRead your workflow skill, then tell me which step would load the administrator's NAV file intoour NAV Records data object. Don't create anything.SparkAgent3 steps1 minute agoReasoningI will load the workflow authoring skill and list the workspace's data objects using `loa…Spark List Data Objects122msWorkflow Authoring5msNAV Records is one of the workspace's 12 data objects. A Read CSV File step reads theadministrator's file into records, and a Save Record step writes them to NAV Records.
A loaded skill shows up in the steps, next to the agent's tool calls.

Where to go next