# perf\(server\): stop caching unused OpenCode tool parts \(\#9738\) · screen/t3code-personal

[View on GitCafe](https://git.cafe/screen/t3code-personal/commit/c8f77e0d441264efb0acfac312e852c81ae3da83)

Repository: [screen/t3code-personal](https://git.cafe/screen/t3code-personal)

Visibility: public

Requested revision: c8f77e0d441264efb0acfac312e852c81ae3da83

Requested commit: c8f77e0d441264efb0acfac312e852c81ae3da83

Commit: c8f77e0d441264efb0acfac312e852c81ae3da83

Tree: 03d09d6070deb71d5b0e633bcf24915f24b2ecd2

Author: Theo Browne

Committer: GitHub

## Message

```
perf(server): stop caching unused OpenCode tool parts (#9738)

Do not retain OpenCode tool parts after emitting their runtime events. The remaining cache readers need text, reasoning, or step-usage parts, not tool input and output.

Keep tool lifecycle events, output bytes, late-role assistant text, and usage handling unchanged. Add focused lifecycle coverage and verify retained memory through the real adapter.

Created with GPT-6 Astra (preview) in Codex.
```

## Parents

- [088cc3f95599b7b933289d719fa1e9b0608cb4cd](https://git.cafe/screen/t3code-personal/commit/088cc3f95599b7b933289d719fa1e9b0608cb4cd?format=markdown)

[Source at this commit](https://git.cafe/screen/t3code-personal/tree/c8f77e0d441264efb0acfac312e852c81ae3da83?format=markdown)