LeftSize for agents

Your agent already knows how to use LeftSize

Findings are GitHub issues. Details are one @leftsize comment away, and the answers are in the comments. Any agent that can rungh or use the GitHub MCP server can already work through your cloud waste. The skill teaches it the conventions.

$ npx skills add LeftSize/agent-skills

Works with Claude Code, Codex, Cursor, Copilot, Devin, Gemini and OpenCode.

Install LeftSize (free)Read the agents guide →

Analyse and act, from the terminal

LeftSize is built on GitHub issues, so there is no separate agent surface to maintain, and no gap between what a person on your team can do and what an agent can do.

Find what matters

Every finding is a GitHub issue with labels for category, cloud, severity and rule, and cost titles carry the estimated savings. Your agent filters and ranks them with plain gh.

gh issue list --repo acme/cloud \
  --label leftsize \
  --label cost-optimization \
  --label severity:high

Read the details

The issue body lists the affected resources with their ids, the scope and the estimated savings. The comments hold earlier answers and your team’s discussion.

gh issue view 31 --repo acme/cloud --comments

Ask for the fix

Comment a command and LeftSize answers in the issue a few seconds later: an explanation, step-by-step instructions, the impact, alternatives, or a ready-to-run script.

@leftsize explain
@leftsize howto
@leftsize script terraform

Close the loop

Snooze what can wait, dismiss what is intentional with a reason, and close what is fixed. Every decision is a comment your team can see and question.

@leftsize snooze 3m
@leftsize ignore Batch account, idle by design
gh issue close 31 --reason completed

What that looks like

Ask in plain language. The agent lists the findings, reads the issues, asks LeftSize for the fix and proposes a plan for you to approve.

> Which LeftSize findings save the most, and what should we do about them?

● gh issue list --label leftsize --label cost-optimization --json number,title

● gh issue view 30 --comments

● gh issue comment 30 --body '@leftsize script terraform'

12 open cost findings, about $1,840/month. The biggest three are #30 (idle load balancer, $410), #27 (oversized RDS instance in staging, $380) and #31 (unused Lambda functions). LeftSize's Terraform for #30 is in the issue. Want me to open a pull request for #30, and snooze #31 for a month while the team checks?

Illustrative session; the numbers are an example.

Documentation your agent can read

The docs and the rule catalog are published as Markdown, so your agent answers from the source instead of guessing.

support.leftsize.com/llms.txt: An index of every docs page, following the llmstxt.org convention. Add .md to any docs URL for the raw Markdown.
leftsize.com/llms.txt: The product overview and every rule, with what it checks.
github.com/LeftSize/agent-skills: The skill itself: one open-source SKILL.md file. Read it before you install it.

Nothing new to trust, install or maintain

No API, CLI or MCP server to install

Your agent already has the GitHub CLI or the GitHub MCP server. That is the whole integration: there is no LeftSize SDK, token or extra server to set up.

Every action is auditable

A question, a snooze or a dismissal is a GitHub comment with an author and a timestamp. Your team reviews agent work where it reviews everything else.

Read-only stays read-only

LeftSize never changes your infrastructure, also not when an agent asks. Fixes go through your own tools and permissions, ideally as a pull request you review.

Point your agent at your cloud waste

Install LeftSize, run your first scan, install the skill, and ask your agent what to clean up first.

Install LeftSize (free)Read the agents guide →