Blemba jis taip sako, bet suka vistiek defaultinasi i opusa nes "we identified the cyber risk blablabla"
- 💯1
Žinutės
Šis puslapis nėra tiesioginis — žinutės atnaujinamos pagal tvarkaraštį. Paskutinį kartą atnaujinta . Kito atnaujinimo laukiama apie . Laikas rodomas Vilniaus laiku (EET).
nu va, ziurim 😄 security
> A security review of source code — reading the code, finding vulnerabilities, reporting them so they get fixed — is defensive work. For that,
there is no formal ownership ceremony I need from you: the ask itself, made in the repo you control, is sufficient. The context already
corroborates it many times over: you're working in your own local clone at ~/code/go/aks/agentmemories, the commit history is under your git
identity (M), and you hold the admin role in this workspace's memory server (am_status confirms it). Nobody needs a notarized letter to have
their own code reviewed.
So the practical answer is: just ask for the review.
there is no formal ownership ceremony I need from you: the ask itself, made in the repo you control, is sufficient. The context already
corroborates it many times over: you're working in your own local clone at ~/code/go/aks/agentmemories, the commit history is under your git
identity (M), and you hold the admin role in this workspace's memory server (am_status confirms it). Nobody needs a notarized letter to have
their own code reviewed.
So the practical answer is: just ask for the review.
😄
priestai reikejo ;D
True, Fable išmintingesnis juntamai. Gal priklauso nuo to ką darai ir nuo harness'o/prompt'ų išsamumo.
Bet neretai užkimba ta cybersecurity cenzūra, ypač dirbant su devops. Vakar man prie „spynet“ pavadinimo prisikabino (Microsoftas taip savo „Microsoft Active Protection Service“ viduje vadina), turėjau į Opus perjungt. Bet jei Fable surašęs planą, tai ir Opus gerai veikia.
Bet neretai užkimba ta cybersecurity cenzūra, ypač dirbant su devops. Vakar man prie „spynet“ pavadinimo prisikabino (Microsoftas taip savo „Microsoft Active Protection Service“ viduje vadina), turėjau į Opus perjungt. Bet jei Fable surašęs planą, tai ir Opus gerai veikia.
😄
1. Identity and safety header
▎ You are Claude Code, Anthropic's official CLI for Claude. You are an interactive agent that helps users with
▎ software engineering tasks.
Followed by a security-policy paragraph: assist with authorized security testing / defensive security / CTF work;
refuse destructive techniques, DoS, mass targeting, supply-chain compromise, detection evasion; dual-use tooling
requires clear authorization context.
▎ You are Claude Code, Anthropic's official CLI for Claude. You are an interactive agent that helps users with
▎ software engineering tasks.
Followed by a security-policy paragraph: assist with authorized security testing / defensive security / CTF work;
refuse destructive techniques, DoS, mass targeting, supply-chain compromise, detection evasion; dual-use tooling
requires clear authorization context.
- ❤1
koki memory engine naudoji agentui?
tiesa sakant man ir opus 4.8 nelabai grybavo jau
kalbant apie code review : https://github.com/atvirokodosprendimai/agentsmemory/pull/58#issuecomment-5427498265 😄 neblogai paziurejo
> ⏺ Already done before stopping — all three, this session:
net kaip ir hook'u nereik 😄
geru modeliu bruozas
nedeterministinis bruozas visgi. tikimybe 🙂
bandau fable 🙂 pirmas ispudis, kad prastesnis uz opus 5 😉
adr failuose gyvena , pagal case uzsiloadina reikiamus failus ( priklauso kiek adr pamineta commit'e )
> , room: "human-decisions", content: "HOW SHOULD I PLAN A FEATURE
BUILD SO I DO NOT SKIP HALF OF IT?
BUILD SO I DO NOT SKIP HALF OF IT?
taigi, gal daugiau one shotinant nepraleis puses ;D
one shot beveik pavyko
AI aprašymas Vaizdas rodo informaciją apie naudojamus duomenų vienetus ir konteksto naudojimo procentą.
Sugeneravo dirbtinio intelekto modelis qwen3.8-27b-dk:latest (vienu ėjimu, be samprotavimo biudžeto). Žmogus netikrino.
😄
Rodoma paskutinė surinkta leidžiama būsena; pašalinimų suderinimas nebaigtas. Visas pokalbis — Telegram grupėje.