{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreidafxydzx27lete3nxqev3arcj5ixqf3bavhvjaumbm3lgnc54q34",
"uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mkc5udu44mm2"
},
"path": "/t/build-a-context-window-manager-in-24h-stick-to-the-norm-or-write-your-own/1379728#post_1",
"publishedAt": "2026-04-25T02:35:43.000Z",
"site": "https://community.openai.com",
"textContent": "Here’s what I did:\n\n1. Built a proxy that intercepts Codex’s calls to OpenAI and rewrites them on the fly.\n\n2. Replayed 3,807 rounds of SWE-bench Verified traces through it: avg prompt 44k → 6k tokens (-87%).\n\n3. Posted it here to get the next reduction applied to my confidence interval — starting with the inevitable “How about accuracy?”\n\nnpx -y pando-proxy",
"title": "Build a context window manager in 24h: stick to the norm or write your own?"
}