I had my claw test with several models and modified the prompt and caching and somehow a grep-based script without LLM at all was quite a strong competitor for most models. Sonnet 4.6 found more verified features, knew the keywords and nuance where a description would talk about non-product stuff etc.
I bet it can be improved but I also have to finish a feature.
LeoWandersleb / Leo Wandersleb
npub1gm…78rf6
2026-02-26 00:04:46 CET
in reply to nevent1q…jzmn
Author Public Key
npub1gm7tuvr9atc6u7q3gevjfeyfyvmrlul4y67k7u7hcxztz67ceexs078rf6Published at
2026-02-26 00:04:46 CETEvent JSON
{
"id": "175d2d0ab312d2cdbd1ca01d972a001a18f79759efe9d2a3e538751ad2819b3f",
"pubkey": "46fcbe3065eaf1ae7811465924e48923363ff3f526bd6f73d7c184b16bd8ce4d",
"created_at": 1772060686,
"kind": 1,
"tags": [
[
"e",
"d8bc5edcf2f1f11e80808e0aa337c65b8851acb1da3a6516fab6a08080f3a6a2",
"wss://relay.primal.net",
"root",
"46fcbe3065eaf1ae7811465924e48923363ff3f526bd6f73d7c184b16bd8ce4d"
],
[
"e",
"3dc6e4b512212ade087fecbd80522a67e547d8f163206038806a67d48dd059a9",
"wss://relay.damus.io/",
"reply",
"d22bcdb1d2505bc060f65a86cd5f20532ec5f3e41c06aea9edac39f6ba76fd6e"
],
[
"p",
"d22bcdb1d2505bc060f65a86cd5f20532ec5f3e41c06aea9edac39f6ba76fd6e"
],
[
"client",
"jumble"
]
],
"content": "I had my claw test with several models and modified the prompt and caching and somehow a grep-based script without LLM at all was quite a strong competitor for most models. Sonnet 4.6 found more verified features, knew the keywords and nuance where a description would talk about non-product stuff etc.\nI bet it can be improved but I also have to finish a feature.",
"sig": "2f6d40e9a20913a416c62a712b34a40ae3cc543a6ffc7149d99d7d5f236323a1db6c2aa9177f29abb7831e665bc3b7fa48a52b63ae2e6560a97c4db7b544507f"
}