Astra agent 反复从零开始烧 token 的解法
作者认为上下文窗口再大也救不了每次重启的流程，主张跨运行保留状态、决策与失败记录，并要求可验证的产出证据。
your GPT-6 Astra agent can burn a million tokens and still be on step one. a bigger context window won’t fix a workflow that keeps starting over. the agent reopens yesterday’s decisions. retries yesterday’s failures. writes another update saying it’s “making progress.” you’re paying for the same work twice. here’s what to preserve between runs: • state - what’s finished, what’s blocked, what happens next • decisions - what was chosen and why • failures - what broke and what must change before retrying then require evidence. a passing test. a working artifact. a verified state change. “almost done” isn’t one of them. I broke down the architecture from persistent memory to activity tracking and verifiable progress. read this before your astra agent spends another hour explaining why it needs another hour ↓
这条是引用着另一条写的：x.com/mr_kozh/status/2097221341107925350