Skip to main content
yertle-sre lets you ask plain-English questions about your software systems — “are any alarms firing for the payments service?”, “what changed in this repo last week?”, “which AWS resources back this Yertle node?” — and get answers grounded in your Yertle architecture graph plus live state from the yertle, aws, and gh CLIs. Repo: model-context/yertle-sre
Status: alpha (v0.1). The CLI surface and configuration are still evolving. Expect changes.

​
How it works

yertle-sre is a small LangGraph agent powered by Anthropic’s Claude. It runs in a ReAct loop: Claude decides which tool to call, the tool returns results, Claude decides what to do next, and the loop ends when Claude has an answer. Three read-only tool runners are exposed to the model:
  • yertle_run — wraps the yertle CLI (orgs, nodes, tree, canvas, about, config) with --format json auto-appended.
  • aws_run — wraps the aws CLI, restricted to read verbs (describe-*, list-*, get-*, show-*, search-*, head-*, lookup-*).
  • gh_run — wraps the gh CLI for list/view/status reads and GET-only gh api calls.
Destructive invocations are refused before they reach the CLI, so the agent is safe to run against production by default.

​
What you can do with it

Interactive REPL that preserves context across questions:
It also ships as a Python library — from yertle_sre import ask, build_agent — for embedding in larger workflows.

​
Setup

Requires Python 3.11+ and the yertle, aws, and gh CLIs already authenticated on your machine. yertle-sre uses your credentials and inherits your shell environment — there’s no separate credential store.
See the yertle-sre README for the full configuration reference (model selection, tool timeouts, iteration caps, LangSmith tracing).

​
Contributing

The repo is open source. Bug reports and PRs welcome — especially around the system prompt and example questions, which directly shape agent behavior.