Octos: a minimalist pipeline orchestrator for CLI LLM agents

Fri, 06 Feb 2026 10:45:00 +0100

I’ve been using CLI agents (kiro, claude, goose…) daily and I got tired of doing the same thing over and over: run a prompt, look at the output, copy what matters, feed it to the next prompt. Repeat. When something breaks at step 5, start over.

So I wrote Octos. It’s a pipeline runner for CLI agents. You describe steps in YAML and it chains them together. That’s basically it.

agent:
 cmd: "kiro-cli"
 args: ["chat", "--no-interactive", "--trust-all-tools"]

steps:
 - name: analyze
 prompt: "Analyze this codebase and list the main issues"
 save_to: analysis.txt

 - name: fix
 load_from: analysis.txt
 prompt: |
 Fix the issues found:
 {{artifact.analysis}}

Any binary that takes a prompt and writes to stdout works. No SDK, no adapters. You can even swap agents between steps — use a cheap model for the boring analysis and an expensive one only for the step that needs it.

Cli on VicenDominguez

Octos: a minimalist pipeline orchestrator for CLI LLM agents