OpenSpec + Ralph = OpenSpec Loop

[viteinfinite]

Hi! Inspired by the community's recent interest in Ralph, I thought it would be more useful to create a loop around OpenSpec instead—via a dedicated skill. Here it is:

🔁 openspec-loop

npx skills add viteinfinite/skills

Benefits

• Clean slate, every time => better context, (hopefully) less token consumption
• No contamination from previous attempts
• Focused attention on a single spec

Suggested Workflow

1. Write your specs
2. Invoke the /openspec loop skill
3. Choose the implementation order
4. Start the loop

Considerations

• Maybe I've simply reinvented the wheel, but I didn't find anything similar on the various skill repositories.
• Most of the implementation details come from obra/superpowers

Feedback welcome!

[dhndmn]

I like this approach! I'm curious if a per-task instead of a per-spec loop would be useful, too. If a spec is comprised of 10 tasks, then address each task individually in order until complete, logging any failures/learnings after each iteration, then wipe the session and address the next task until the full tasks.md has been completed. Overkill? Do you find most context windows sufficient to handle full specs with one session? Or is the key to write smaller specs?

[v-karbovnichy]

I tried this one, but it kinda complex thing with shenanigans
https://github.com/wenqingyu/ralphy-openspec

[TabishB]

Not a Ralph loop, but agent teams in Claude code work decently as well. All you have to do is enable the experimental feature and tell Claude to use "agent teams" when applying.

/opsx:apply use agents team to implement the following change proposal:
simplify-skill-installation