Skip to main content
On this page

AI

Skills and workflows

Every skill and workflow Kater ships, the agent that runs each one, and what goes in a skill file and a workflow folder.

Audience
Business teams and data teams

A skill is how your agent does one task. A workflow is a process in steps, with one agent for each step. An agent is the role that does the work. Each is a Markdown file in ai_tools/, and each can be a starter that anyone runs by name.

Agents, skills and workflows

AgentSkillWorkflow
What it holdsA role: a persona, the skills it owns and its rulesThe procedure for one taskA process in steps, and the agent for each step
Where it isagents/<name>.mdskills/<name>.mdworkflows/<name>/workflow.md, with one file for each step in steps/
What it remembersNothingNothingEach step it completes, in your brainstorm folder, so it can carry on later
A starterWith a starter: blockWith a starter: blockWith a starter: block
On a scheduleNoNoWith a schedule: block

Keep each kind to its job: the persona in an agent, the procedure in a skill or a workflow step, and the state in the brainstorm folder.

Every skill

Kater ships 28 skills. A skill with a starter shows on AI Tools › Starters. The others are steps that an agent or a workflow uses.

SkillCategoryAgentStarterWhat it does
resolve-workspacerouteConductorFinds the one workspace a request needs, or makes a personal one after your yes
planningplanPlannerPlan a projectTurns a goal that takes more than one change into a plan in phases
run-planplanConductorRuns a plan’s phases one at a time. The Developer builds each one, and QA checks it.
onboardplanPlannerOnboard meA first chat that fills in your about.md and ends on one of your own numbers
about-meplanPlannerUpdate About youRecords one thing about you under the right heading of your about.md
digestplanPlannerWeekly digestDrafts a digest of the metrics you watch, on a schedule you approve
watchplanPlannerWatch a metricDrafts a watch on one metric, with a line profiled from its history
schedule-workflowplanPlannerSchedule a workflowSets up, changes, pauses or stops a scheduled workflow
touranalyzeAnalystTour my dataWhat the connection holds, and three questions it answers now
brief-meanalyzeAnalystBrief meWhat moved in the metrics you watch since your last brief
compareanalyzeAnalystCompare two periodsOne metric in two periods, and the dimensions that moved it most
explainanalyzeAnalystExplain a metricHow a metric, or a number on screen, is computed, in plain words
build-plananalyzePlannerThe plan for one question in Why did it change?
build-analysis-daganalyzePlannerTurns that plan into queries, decisions and follow-up checks, in order
profile-parametersanalyzeAnalystProfiles the data for a value the plan needs, such as a threshold
select-analysis-parametersanalyzeAnalystPicks that value, and records why
run-analysis-daganalyzeAnalystRuns the plan’s queries in order
diagnose-findingsanalyzeAnalystRuns the follow-up checks the plan declared
query-authoringbuildDeveloperThe rules for every query and workspace root
configure-workspacebuildDeveloperPicks the model a workspace uses and adds what its queries need
run-phasebuildDeveloperBuilds one phase of a plan
build-app-uibuildDeveloperBuild an appBuilds or changes an app
model-tablebuildDeveloperModel a new tableTurns synced warehouse tables into shared models and first queries
new-skillbuildDeveloperMake your own commandWrites a skill of your own, which you start by name
checkreviewQACheck a numberTraces one number back to its source
test-modelreviewQATest a modelLocks in answers you trust as tests of a shared model
review-analysisreviewQAReview an analysisChecks a finished analysis before its findings are written
synthesize-findingsreportReporterWrites the findings and recommendations from reviewed evidence

Every workflow

Kater ships five workflows.

WorkflowStarterStepsWhat it does
deep-analysisWhy did it change?7Finds a workspace, then plans, builds, profiles, runs, reviews and reports on one question
build-query-contractDefine a metric4Finds what exists, starts a draft, changes it until it is right, then checks and publishes one query
resolve-workspaceStart a workspace4Lists the workspaces you can read, rates them for the request, then picks one or makes one and checks it
calibrate-analysis-parameters2Profiles an approved analysis again and recommends a new value. It never changes the value itself.
monday-briefScheduledIn one fileEvery Monday at 08:00, briefs you on what moved last week. See Scheduled workflows.
The Workflows tab with Deep Analysis open, showing its goal and its step map table, and Monday Brief in the list with Every Monday at 08:00 under its name
AI Tools › Workflows, with Deep Analysis open on its step map.

What is in a skill

A skill is one file: a header in YAML, then the procedure in Markdown. This is the header of brief-me:

connections/snowflake_prod_connection/ai_tools/skills/brief-me.md
---
label: Brief Me
name: brief-me
tags: kater-default
category: analyze
owned_by: analyst
description: Brief the person on what moved in the metrics their about.md names, from verified definitions, since a date they give or their last brief.
starter:
title: Brief me
summary: What moved since last time
icon: newspaper
group: Understand
order: 1
ask: Brief me on what moved in the metrics I watch since my last brief.
inputs:
- name: since
description: How far back to look, such as last Monday or the start of the month. Left out, since your last brief, or a week.
---
KeyWhat it holds
labelIts name in the library
nameHow you start it, such as /brief-me
tagskater-default on each file Kater ships
categoryroute, plan, analyze, build, review or report
owned_byThe agent that runs it
descriptionOne sentence. Your agent reads it to decide when to use the skill, so say when it applies.
guideThe guide in guides/ it follows, when it has one
invokesOther skills it hands over to
starterMakes it a starter: its title, a one-line summary, an icon, its group and order, the words it starts with (ask), and the inputs a person can give it

The body says when to use the skill, the steps it follows, and when to stop:

# Skill: brief-me
## When to use
Use this when the person asks what changed, how things are going, or for their brief.
To find out why something moved, hand over to deep-analysis.
## Procedure
1. **Read what they watch:** `## Metrics I watch` in `{user_root}/about.md`.
2. **Set the window.** From `since` or their words.
3. **Find each metric's verified definition.**
...
## Stop conditions
- No metric has a verified definition: say which lack one, and offer build-query-contract.

Words in braces, such as {user_root}, come from config.yaml. When a starter reaches your agent through the Kater MCP, Kater fills in your own email, so {user_root} is your own folder.

What is in a workflow

A workflow is a folder. workflow.md holds its header, its goal and a step map. Each step is a file in steps/, and your agent reads only the one it is on.

workflows/deep-analysis/
workflow.md
steps/
step-00-resolve-workspace.md
step-01-plan.md
step-02-build.md
step-03-profile-and-calibrate.md
step-04-execute-and-diagnose.md
step-05-review.md
step-06-report.md
connections/snowflake_prod_connection/ai_tools/workflows/deep-analysis/workflow.md
---
label: Deep Analysis
name: deep-analysis
tags: kater-default
description: Resolve a workspace, plan, build, profile, execute, review, and report one complex governed analysis.
config: '../../config.yaml'
engine: '../../engine.md'
workspaceMode: resolve
outputFile: '{brainstorm_folder}/{{analysis_slug}}.md'
starter:
title: Why did it change?
summary: Trace a change to its drivers
icon: trending-up
group: Understand
order: 2
ask: Why did
---
KeyWhat it holds
configThe config it loads first
engineThe rules it runs by: one step at a time, with a checkpoint after each
workspaceModeresolve finds a workspace first, required needs one already picked, and none uses no workspace
outputFileThe note it writes in your brainstorm folder
starterMakes it a starter
scheduleMakes it a scheduled workflow, such as cron: "0 8 * * 1" for every Monday at 08:00

The step map in workflow.md names each step’s file, its agent, the skills it uses and what it produces. To change who does a step, change its agent there.

Next steps