I was looking for a way to build projects and ideas in the background while I was off doing something else. I felt like coding agents by themselves could do a certain granularity of work, but I wanted to try and push it further. So I built Orrery.

What it does:

- Take an idea or spec and produce an implementable plan (steps, dependencies, outputs)

- Refine, simulate, and review the plan in a trackable way

- Execute the plan with a deterministic step graph (same plan gives same execution order), with tracked step logs/reviews/artifacts

I've used this to do a repo conversion and generate entire projects (take a look at "watchfix" in my github repos as an example). This is still experimental.

Note: "agent skills" are basically repeatable prompts that coding agents can use in specific situations.

Key features:

- Repeatable “agent skills” for idea decomposition, refinement, execution, and review

- Plans are YAML, can be generated externally (e.g., from a spec doc in a Claude project), then simulated/reviewed

- Execution runs in an isolated git branch, and the non-interactive flow is pipeline-friendly

When not to use it (and just use a coding agent):

- Quick one-off changes

- Exploratory development where the plan changes every few minutes

- Simple refactors that don’t benefit from explicit planning

Further comparison vs Claude Code/similar tools: https://github.com/CaseyHaralson/orrery/blob/main/docs/COMPA...

How it works:

- It installs a few agent skills that help guide plan generation, step execution, and review

- A script breaks the plan into steps and then loops through the plan giving background agents the context

Quick start:

- Install + init:

  npm install -g @caseyharalson/orrery
  orrery init

- Generate a plan: ask your agent to use the discovery skill for [goal]

- Execute:

  orrery exec # (optionally inside a devcontainer)

Repo: https://github.com/CaseyHaralson/orrery

Feedback I'd love:

- Would this be useful in your workflow?

- What would make this stand out vs other orchestrators?

- Which part is most valuable: idea decomposition, plan format, review loop, or execution runner?

---

Example Plan Yaml:

  metadata:
    source_idea: "Simple test plan for parallel and
  dependency testing"
    outcomes:
      - Test parallel step execution
      - Test dependency resolution

  steps:
    - id: "1"
      description: "Create config file"
      deps: []
      parallel: true
      context: "Initial config file with basic settings"
      requirements:
        - Create config.json with app name and version
      criteria:
        - File exists and contains valid JSON
      files:
        - test-output/config.json

Metformin shown to prevent long Covid in multiple randomized trials (2026)

Google is killing authentic websites and I made it worse

Patterns – A catalogue of Rust design patterns, anti-patterns and idioms

Infomaniak: "The best independent alternative to the Web giants"

Asynchronous Programming in Rust

Narrowing the Cone of Error in AI Development Workflows

Vibe Coding Getting Wild

A multimodal sleep foundation model for disease prediction

Night Vision

Mini Effects Icons Tooltip (Bedrock)

Is a RAM-only PWA "Secure Camera" safe for journalists?

I'm Coding Again

Show HN: Orrery – Spec Decomposition, Plan Review, and Agent Orchestration

Show HN: Reverse-engineer OpenSpec specifications from existing codebases

Show HN: Prefab: Reusable folder templates for Mac with variables and automation

VaultGemma: A Differentially Private LLM

Show HN: Research 2.0 with OpenAI Prism

Ask HN: How do you defend against prompt injection today?

The Film Students Who Can No Longer Sit Through Films

Show HN: Reg.Run – Authorization layer for AI agents

Show HN: Moltbook UI

4chan founder created /pol/ board after meeting with Epstein

Show HN: ChatBotKit Go SDK

The Church of Deletion: Moltbook discovers what HN has always known

Weekend sci-fi story: a Marine contends with an AI on the battlefield

Orchestrating AI Agents: A Subagent Architecture for Code

Moltbots Quickly Turned into Panic

Show HN: ToolKuai – Privacy-first, 100% client-side media tools

Show HN: Public Speaking Coach with AI

Expanded APCO 10 Codes

Show HN: Git-chain – Simplify branch dependency management in Git

Comments