Coding Agents

LeftSize has no separate API, CLI or MCP server for agents, and it doesn’t need one. Findings are GitHub Issues, you ask for detail with an @leftsize comment, and the answer arrives as a comment.

That means any agent that can run gh, or use the GitHub MCP server, can already work with your findings. This page covers the conventions it needs to know.

npx skills add LeftSize/agent-skills

The mental model

Everything is an ordinary GitHub action, so it shows up in the issue history with an author and a timestamp, visible to your whole team.

Prerequisites

• The GitHub CLI installed and authenticated as yourself (gh auth login), or the GitHub MCP server
• Access to the repository where LeftSize opens issues

Finding the findings

Every LeftSize issue has the leftsize label and is opened by leftsize[bot].

# All open findings
gh issue list --repo acme/cloud --label leftsize

# Cost findings on AWS
gh issue list --repo acme/cloud --label leftsize --label cost-optimization --label cloud:aws

# Only high severity
gh issue list --repo acme/cloud --label leftsize --label severity:high

Cost issue titles end with the estimated savings, such as [COST] 3 unattached disks (≈ $42/mo), so an agent can total and rank them from gh issue list --json number,title,labels.

Reading a finding

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

Each issue covers one rule in one scope (an Azure resource group, or an AWS account and region). The body lists the affected resources with their ids, the estimated savings, the available commands and example fix commands. Every scan rewrites the body with the current state. The comments hold earlier answers from LeftSize and your team’s discussion.

Asking for more detail

Post a command as a comment. LeftSize replies with a new comment, usually within a few seconds:

gh issue comment 31 --repo acme/cloud --body '@leftsize script aws-cli'
gh issue view 31 --repo acme/cloud --comments

explain, howto, impact, alternatives, script <type> and prompt only add a comment and change nothing. See Commands for the full reference.

Acting on a finding

# Postpone: closes now, reopens after 3 months if the problem is still there
gh issue comment 31 --repo acme/cloud --body '@leftsize snooze 3m'

# Dismiss for good, with a reason for the team
gh issue comment 31 --repo acme/cloud --body '@leftsize ignore Batch account, idle by design'

# Fixed it
gh issue close 31 --repo acme/cloud --reason completed --comment 'Removed the unused functions.'

LeftSize itself only reads. It never changes your infrastructure, and commands like stop or delete are refused. When the agent fixes something, it does so with your own tools and permissions, ideally as a pull request to your infrastructure code.

Things agents get wrong

• Any @leftsize in a comment is a command. LeftSize runs the first @leftsize anywhere in the comment. An agent writing a note should write “LeftSize” without the @.
• Comments from bots are ignored. Commands posted by github-actions[bot] or another GitHub App do nothing. The agent must comment as you.
• Replies are asynchronous. The answer isn’t in the output of gh issue comment; read the comments again a few seconds later.
• script defaults to Azure CLI, also on AWS issues. Use @leftsize script aws-cli for AWS.

Running a scan

gh workflow run leftsize.yml --repo acme/cloud

The run’s job summary shows the findings per rule and the estimated savings. See Troubleshooting if a scan fails.

Giving agents the documentation

These docs are available as Markdown for agents:

Which agents work

Anything that can run shell commands or use the GitHub MCP server: Claude Code, Codex, Cursor, Copilot, Devin, Gemini, OpenCode and others. Nothing here is agent-specific.

Related

• Commands: the complete command reference
• How It Works: the scan cycle and issue lifecycle
• Rules: every rule LeftSize checks