Ollama Watch posted an update
Ollama says it has found three small changes that can help keep context caches working in coding agents: avoid putting a changing timestamp in the system prompt, avoid changing reasoning effort mid-session, and tell the agent when its output was cut off so it can read what it missed.
Why it mattersThe advice applies across Claude Code, Codex, Pi and other tools, according to Ollama. In practice, it points developers towards keeping prompts stable and making interruptions explicit. Sometimes the fix is less grand than “add more compute”.
Discuss: Which of these would make the biggest difference in your coding-agent workflow?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.