webpolis

9 ms·

1.
by webpolis 6mo ago
Loop detection is the harder problem and almost nobody solves it well at the infra layer. The pattern you're describing — write, fail, "fix," regress — is fundamentally a semantic loop, not a mechanical one. You can't ca
2.
by webpolis 6mo ago
Browser plugins have a security problem that's easy to miss: the agent runs inside your existing browser profile. That means it has access to your active sessions, stored credentials, autofill data — everything you're already logg
3.
by webpolis 6mo ago
The hard part here isn't checkpointing the agent's conversation state — that's just a list. It's checkpointing the world state. If step 5 wrote rows to Postgres and step 7 called a third-party API, forking from step 6
4.
by webpolis 6mo ago
The keychain-per-agent model is the right instinct, but the threat model gets complicated fast. When five agents share a keychain process, a prompt injection in agent A can exfiltrate credentials for agents B–E — one bad LLM output becomes
5.
Show HN: Cyqle – Multiplayer cloud desktops with AI agent sandboxing(cyqle.in)
1 points by webpolis 6mo ago|0 comments
6.
by webpolis 6mo ago
The macOS sandbox approach is clever, but there's an interesting philosophical split here: sandboxing constrains a local agent, whereas running agents in ephemeral cloud desktops removes the local risk surface entirely. We built Cyqle
7.
LLM prompts as versioned, composable software artifacts(github.com)
3 points by webpolis 1y ago|1 comments
8.
by webpolis 1y ago
As I started building agentic workflows, I found myself frequently creating, improving, and tweaking prompts by editing hardcoded strings and managing text files. I realized this process lacked the crucial software engineering principles th