Summary
Instead of losing tool output when it exceeds the model context, the SDK stores the complete result in a per-session cache and exposes a cline://cache URI. Agents can retrieve the remaining content by line range, with session-level size and expiry limits.
Why it matters
Large tool responses are a common reliability failure in agent workflows; this change preserves evidence without injecting the entire payload into every model turn.
Related coverage
- GitHub Copilot CLI adds GPT-6.1 Sol and tighter session controls
- VS Code 1.140 makes multi-agent development a first-class workflow
- Claude Code 2.1.286 improves permission and cloud-credential recovery