Skills Builder - LumiOps

The building blocks every agent is made of.

Skills are structured recipes — steps, tool calls, constraints and an output schema. Lumi routes every question through the right ones before a single tool fires. Browse the catalog, clone one, or write your own from an SOP.

built-in skills

domains covered

routing checks per query

⇄Skill Routeridle

›

CANDIDATE SKILLS32 in catalog

✓

3 skills routed → tools invokeddb-pool-diagnostics · latency-correlation · deploy-timeline

MatchCheckRouteInvoke

SKILL ROUTER

Four checks before any tool fires.

Routing isn’t just similarity search. Every candidate skill is tested for relevance,
whether its connector is live, whether the skill is enabled, and whether you’re allowed to
run it.

embedding

Relevance

Semantic match between the question and the skill’s described purpose.

connector

Availability

Is the system this skill depends on actually connected right now?

state

Skill state

Enabled, drafted, deprecated — only live skills get routed.

RBAC

Permission

Your role decides what may run. Denied skills are dropped, not hidden.

BUILT-IN SKILLS CATALOG

Organized by domain. Filtered by
persona.

Compute, network, storage, cloud, database, identity, security, observability and
services — with a marketplace and your own workspace skills alongside.

⊙ Add Memory Leak Detection skill

Workspace saw memory-leak patterns trigger 14× last week. The closest existing skill is K8s Pod Diagnostics (76% acceptance on this pattern) — a dedicated skill from the marketplace would handle it natively.

Marketplace → install "Memory Leak Detection"Expected acceptance +12pts on memory-leak incidents

⚠ Refresh AWS Investigation few-shot examples

Calibration drifting 14 points over 30 days. Newer AWS SKUs (Graviton4, Tm2) are absent from the current prompt.

Open skill · Detail → Versions → Promote v4 (drafted)Expected calibration 74% → 86%

27of 32 skills available PERSONA SRE / DevOps

⌕

No skills match that search.

WHAT A SKILL ACTUALLY IS

A recipe, not a prompt.

1

Steps & procedure

The ordered method an expert would follow, written down once.

2

Tool calls

Exactly which systems it may query, and with what parameters.

3

Constraints

What it must not do — the guardrails travel inside the skill.

4

Output schema

A predictable shape, so other skills and workflows can consume it.

5

Connectors

The integrations this skill is bound to, checked before routing.

db-pool-diagnostics.skillv2.3 · enabled

domain: database
persona: [sre, dba]
connectors: [postgres, datadog]
autonomy: discover
steps:
  - pull active sessions + wait events
  - correlate pool saturation vs latency
  - rank blocking queries by impact
constraints:
  # never terminate a session
  - read_only: true
  - max_rows: 5000
output:
  root_cause, evidence[], confidence, next_actions[]

SKILLS GOVERNANCE & GUARDRAILS

Every skill declares its own
autonomy.

Classification lives on the skill itself, so an agent can never exceed what its skills allow.

discover

Read only

Investigates and reports. Cannot change a thing, safe on any site.

manage

Scoped change

Acts on tagged targets within a blast radius, with approval where required.

automate

Unattended

Runs on its own once its acceptance rate has earned the promotion.

MARKETPLACE & CUSTOMIZATION

Three places a skill comes from.

COMPANY CATALOG

Built in by UnityOne

Production-tested skills across every domain, versioned and maintained for you.

32 skills

WORKSPACE

Yours and your team’s

Clone a built-in skill and tune it, or write a new one from your own SOP.

2 in this workspace

MARKETPLACE

Installed on demand

Add specialist skills when Lumi spots a gap — recommended, not guessed at.

6 available to install

CAPABILITIES

Everything skills give you.

▤

Built-in catalog

32 composable skills organized across ten operational domains.

👤

Persona-based

Filter to SRE / DevOps, DBA, NOC or platform — see only what fits the role.

✦

AI recommendations

Lumi spots missing skills and drifting calibration, and proposes the fix.

⧉

Cloning & customization

Fork any skill, edit its steps and constraints, keep the original intact.

⑂

Versioning

Draft, promote and roll back skill versions with calibration tracked per version.

⇄

Skill Router

Test which skills a query selects, and why — before it ever runs.

🔌

Connectors

Each skill declares its integrations; unavailable ones are routed around.

🛡

Guardrails

Constraints live in the skill, so it can only ever do authorized work.

✎

Custom from SOPs

Turn your own runbooks into skills your whole team can direct.

WHY IT MATTERS

The same work, done faster — and
cheaper.

Skills turn one engineer’s best method into something the whole team can run, every
time, without waiting for the person who knows.

MEAN TIME TO RESOLVE

4 hours

40–60%
lower

Routing skips the search phase — the right method runs first, not fourth.

ALERT NOISE

900 alerts/day

~90%
less

Correlation skills collapse floods into a handful of real incidents.

TIME ON FIREFIGHTING

⅓ of the week

back to building

Repeatable diagnosis stops eating the roadmap.

NEW-HIRE RAMP

3–6 months

day
one

A junior directing a skill follows the same steps as your best senior.

👤

For the ITOps engineer

DAY TO DAY

✓

Stop starting from zero

Every investigation begins with an expert’s method already loaded, not a blank terminal at 3am.

✓

Fewer escalations to you

When the skill encodes your runbook, the on-call engineer doesn’t need to phone the one person who knows.

✓

Do the interesting work

Hand the tenth identical CrashLoop to a skill and get your afternoon back for architecture.

✓

Confidence to act

Constraints and RBAC are enforced by the skill, so acting fast doesn’t mean acting recklessly.

◈

For the organization

ON THE BALANCE SHEET

✓

Downtime costs less

Over 90% of enterprises put an hour of outage above $300K. Cutting MTTR by half takes that straight off the top.

✓

Capacity without headcount

The same team covers more estate, because the repeatable half of the work no longer needs a person.

✓

Expertise stops walking out

When your best engineer leaves, their method stays behind as a versioned skill.

✓

Consistent, auditable operations

Same procedure every time, with a full record — which is what auditors and regulators actually ask for.

Ranges reflect published AIOps and ITOps benchmarks — 40–60% MTTR reduction, up to 90%+ alert-noise reduction, roughly a third of engineering time lost to firefighting, and $300K+ as the typical cost of one hour of enterprise downtime (ITIC, Forrester, New Relic, 2024–2026). Your results depend on your estate and starting point.

Get started

Turn your best
runbook into a skill

Start from a built-in skill, clone it, and make it match how your
team actually works.

embedding

connector

state

RBAC

1

2

3

4

5

discover

manage

automate

Build

Orchestrate

Cowork

Resolve

About Us

About Us

Product

Coworker

Orchestrator

Product

Orchestrator

Coworker

Build

Platforms

Estate Models

Agents

Observability

Service Desk