MagnusMagi/check-devops

One command. 8 parallel agents. 312 checks across security, databases, DevOps, observability,
distributed systems, payments, compliance and AI. Every verdict comes with evidence: a file:line or a command output.
## Why
AI makes it easy to ship an app in a weekend, and just as easy to ship it without rate limiting, backups you have ever restored,
CSRF protection, graceful shutdown, or a single alert. Nobody forgets these on purpose. There are simply too many of them to
keep in your head.
check-devops turns that list into an audit. It reads your repository the way a senior backend/SRE reviewer would, item by
item, and tells you what is there, what is weak, what is missing, and what does not apply to your project, with proof.
🔎 Evidence or nothing. No ✅ without a file:line or a command result. If it can't find something it says unverified, not missing.
🛡️ Read-only. It never edits your code. You get a report and choose what to fix.
⚡ Fast and parallel. 8 Explore subagents scan ~39 items each at the same time.
🧭 Prioritized. Security and data-loss gaps first, each with the smallest fix that closes it.
🏗️ Infra-aware. Items covered outside the repo (reverse proxy, firewall, backups, uptime monitors) count when your notes say so.
🔁 Re-runnable. The full table lands in docs/check-devops.md; each new run diffs against the last one, so you can watch the score move.
🌍 Any stack. Node, Python, Go, Rust, Java, Ruby, PHP, .NET, mobile, monorepos. The report is written in your language.
## Quick start
### Option 1: install as a Claude Code plugin (recommended)
Inside Claude Code:
text
/plugin marketplace add MagnusMagi/check-devops
/plugin install check-devops@check-devops
### Option 2: install as a plain skill
bash
git clone https://github.com/MagnusMagi/check-devops.git
cp -r check-devops/skills/check-devops ~/.claude/skills/
For a single project only, copy it into /.claude/skills/ instead.
### Run it
Open Claude Code in your project and type:
text
/check-devops
or just ask: "run check-devops on this repo". (Installed as a plugin, the command is /check-devops:check-devops.)
## How it works
mermaid
flowchart LR
A["/check-devops"] --> B["1 · Project profile stack, run & deploy, CI, IaC, infra notes"]
B --> C1["Agent 1 items 1–39"]
B --> C2["Agent 2 40–78"]
B --> C3["…"]
B --> C8["Agent 8 274–312"]
C1 --> D["3 · Merge"]
C2 --> D
C3 --> D
C8 --> D
D --> E["Priority gaps in the terminal"]
D --> F["Full 312-row table docs/check-devops.md"]
1. Profile. Claude reads what defines the project: CLAUDE.md, package.json / pyproject.toml / go.mod / Cargo.toml,
Dockerfile, compose files, CI workflows, IaC. If a machine-level note exists (for example ~/CLAUDE.md), server-side
items such as the reverse proxy, firewall and backups are counted as covered by infrastructure.
2. Parallel scan. Eight Explore subagents (Sonnet, to keep it cheap) each take a slice of the checklist and return one line
per item: No | Status | Evidence | Note.
3. Merge and report. You get a summary and a prioritized gap list in the terminal; the full table is written to disk.
If the repo has a graphify knowledge graph (graphify-out/graph.json), agents query the graph first and only open the lines it points to, which makes large repos much cheaper to scan.