distill and compress tool error

[reesechou2014]

[reesechou2014]

use v2.0.0

[reesechou2014]

always compress when failed, try still and not stop

[Tarquinen]

Hmm I didn't consider this but maybe the compress tool isn't good with non english languages... I would recommend just disabling it if it happens this much, i'll try to give it some more guidance in the prompt but can only go so far with that.

Is the distill tool failing frequently as well?

[scflode]

This also happens to me using English. I'm on OpenCode 1.1.53.

[reesechou2014]

Hmm I didn't consider this but maybe the compress tool isn't good with non english languages... I would recommend just disabling it if it happens this much, i'll try to give it some more guidance in the prompt but can only go so far with that.

Is the distill tool failing frequently as well? sometimes but not frequently like compress

[Jonathanm10]

same here (english language), happens mostly on compress

[Tarquinen]

ok good to know, i'll look into making the validation less strict

[Tarquinen]

@scflode @Jonathanm10 what models were you guys using?

[Tarquinen]

Compress tool should be significantly improved in latest version (#361), it's disabled by default now as I'm not fully confident in how well it works yet, you can enable it in dcp.json

"tools": { "compress": { "permission": "allow" }}

Please let me know if you test it how it works for you

[scflode]

I used different models. Kimi K2.5, Claude Sonnet 4.5, GPT 5.2 codex. This happened mainly in plan mode.

[Tarquinen]

Anyone have any issues since 2.0.2 went out 2 days ago?

[reesechou2014]

The "compress" tool is already working fine, but "distill" still frequently occurs (especially after using oh my opencode delegate_task for multi-round conversations).
The original conversation has been cleared by me; when I encounter the issue again, I will share my problematic session.

[mrjared16]

Anyone have any issues since 2.0.2 went out 2 days ago?

I still experience the "endstring not found in the conversation"

[reesechou2014]

still have this problems

[Tarquinen]

I still experience the "endstring not found in the conversation"

This is just model performance, if it chooses a compression range that doesn't exist in the content there's nothing I can do to fix that. There's even already fuzzy matching in place in case the model is off by a few characters...

Interesting stuff with the oh my opencode failures though. I'm not really sure how that plugin works, are those background_output non subagent sessions?