O-ALL-WANT

English | 中文 | 日本語 | 한국어 | Deutsch | Français | Español

Why choose when you can have it all? Kids choose. Builders want the whole stack.

20-second pitch

OAW installs a small, local harness into your repo:

• Context router: CLAUDE.md tells the agent which lane to read instead of loading every doc.
• Durable memory: .agents/memory.md carries decisions, bugs, and findings across sessions.
• Compiled wiki: docs/raw/ notes become searchable docs/knowledge/ pages.
• Repeatable skills: .agents/skills/ keeps recurring workflows from becoming random prompt improv.

Measured result: lane routing loads ~3k tokens per task lane instead of ~22.6k for the whole project — 86-87% less startup context.

Read CLAUDE.md first, then AI_CONTEXT.md. I'm building [describe your project]. Use OAW's architecture as the harness, map this project's real state into AI_CONTEXT.md, then tell me which repeated workflows belong in .agents/skills/.

Why are you here?

This is a harness for unapologetically greedy agentic coders.

If you hop between AI coding platforms and treat token efficiency like a competitive sport, you probably know the heartbreak: the chat gets long, the agent gets stupid, the session resets, and suddenly you are paying premium tokens just to re-explain the repo before any real work happens. You do not even reach the hard part before seeing You have hit your limit.

OAW is my answer to that waste.