Skills

Skills are short, structured runbooks the SRI Agent can invoke on demand. A skill turns “the way we do post-deploy verification on this team” or “the standard weekly cost review” into something the agent runs consistently, every time, for everyone. Each skill uses the open Agent Skills format initiated by Anthropic and adopted across the industry (Cursor, Claude Code, GitHub Copilot). A skill written for RubixKube is portable to those tools, and vice versa.

Why skills

Consistent execution

The same runbook, every time, regardless of who is on call.

Narrow tool surface

Each skill declares exactly which tools it can use. Anything not listed is blocked.

Shared institutional knowledge

Tribal knowledge becomes shared capability. New team members inherit it on day one.

Portable format

Same file runs in Cursor, Claude Code, Copilot, or RubixKube.

Anatomy of a skill

A skill is a YAML block plus plain-English instructions. Minimal shape:

name: post-deploy-verification
scope: tenant
agent: sre-agent
description: Compare error rate, latency, and resource usage for a service against the last stable window.
allowed_tools:
  - fetch_infrastructure_snapshot
  - fetch_deployment_history
  - analyze_service_health
instructions: |
  1. Take a baseline from the last 30 minutes of the previous stable deployment.
  2. Take the current window (10 minutes after the new rollout).
  3. Compare error rate, p95 latency, CPU, and memory.
  4. Flag any metric worse than 20% against baseline.
  5. Summarise in a short Slack-ready paragraph.

Field	Purpose
`name`	Unique handle, kebab-case. Used when invoking with `/skills`.
`scope`	`system` (RubixKube-maintained) or `tenant` (your workspace).
`agent`	Which agent can run this. Today only the SRI Agent is supported.
`description`	One sentence. Used in the catalogue and for intent matching.
`allowed_tools`	Explicit tool allowlist. Anything not listed is blocked.
`instructions`	Plain-English runbook. Numbered steps work best.

System skills and custom skills

System skills

Shipped and maintained by RubixKube. Available in every workspace.Current catalogue includes kubernetes-cluster-triage, linear-tickets-to-notion, and research-devops-sre-infra. See the full list in the Agent Skills Store.

Custom (tenant) skills

Written by your team for your workflows. Scoped to your workspace. Versioned per save. Importable from any tool that speaks the open Agent Skills format.Walkthrough: How to add custom agent skills.

How a skill is invoked

Intent match

A user asks a question in Chat. The SRI Agent compares the question against enabled skill descriptions. If there is a confident match, the skill runs automatically.

Explicit call

A user types /skills skill-name in Chat (console or CLI). The skill runs against the current context.

Scheduled run

Some skills are attached to schedules (daily handoff, weekly review). They run without manual invocation and post their output to the configured channel.

Safety boundaries

Skills are not a bypass for Guardian policies. Any tool a skill calls that mutates state still requires the normal approval flow. The allowed_tools list tightens the safety surface further by preventing the agent from reaching for tools outside the list, even if a user’s follow-up would otherwise lead there.

Design principles for good skills

One clear job

Post-deploy verification, on-call handoff, cost review. Not “general ops”.

Numbered steps

Write instructions like a senior engineer’s handover note. Imperative, ordered.

Specific thresholds

“20% worse than baseline” beats “significantly worse”. The agent executes the number.

Narrow tool list

Fewer tools mean safer, more predictable runs.

Common questions

Which agent runs skills?

The SRI Agent. It is the conversational agent that powers Chat. Support for other agents (Remediation, Memory) is planned.

Do skills count against my investigation quota?

A skill run counts as one investigation on your plan. Scheduled skills (daily handoff, weekly cost review) are usually the best use of the quota because they replace work that would otherwise be ad-hoc.

Can I port a skill from Cursor or Claude Code?

Yes. The format is the same. Upload your SKILL.md, map any tool names that differ, and it runs.

How are skills versioned?

Every save creates a revision inside the console. You can roll back a skill from its detail page. Git-based versioning via the RubixKube CLI is on the roadmap.

Agent Skills Store

Browse system skills and manage your custom catalogue.

How to add custom agent skills

Step-by-step guide to writing and shipping your first skill.

Documentation

​Why skills