[BUG] Compactation Fails Agent forgets everything

[dhruvkej9]

Description

The /compact and auto compact fails.

Plugins

"oh-my-opencode@beta",     "opencode-antigravity-auth@beta",     "@plannotator/opencode@beta",     "@franlol/opencode-md-table-formatter@0.0.3",     "@tarquinen/opencode-dcp@latest",     "@mohak34/opencode-notifier@latest"

OpenCode version

1.1.12

Steps to reproduce

1. Run Sisyphus agent with Claude opus 4.5 High think (Antigravity)
2. Manual/Auto Compact fails when context is near to full (90% in my case)

Screenshot and/or share link

Operating System

WSL Ubuntu 24.04

Terminal

Windows Terminal

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Compaction Failing to Run on Time #6286: Compaction Failing to Run on Time
• [BUG]: Auto compaction not triggering reliably with Anthropic models #6068: Auto compaction not triggering reliably with Anthropic models
• Auto-compaction enabled by default, but context_length_exceeded errors still occur in agent workflows #8089: Auto-compaction enabled by default, but context_length_exceeded errors still occur in agent workflows
• Session automatically compacted, destroying the entire working context #2945: Session automatically compacted, destroying the entire working context
• Model in BUILD mode does not have enough context to continue after compaction #3031: Model in BUILD mode does not have enough context to continue after compaction
• compact causes llm to forget convo #2234: compact causes llm to forget convo
• Session Todo-list is forgotten after compaction #5934: Session Todo-list is forgotten after compaction
• Auto-compaction causes the subagent to lose its original context, possibly the system prompt as well #6535: Auto-compaction causes the subagent to lose its original context, possibly the system prompt as well
• Model selection is "lost" on continuation and Agent selection is "lost" on compaction #8349: Model selection is "lost" on continuation and Agent selection is "lost" on compaction

Feel free to ignore if none of these address your specific case.

[super-unique-ai]

Weird, my OpenCode is smart enough to clear drop cache of context to recover memory and continues to work. I am using desktop OpenCode. You should probably upgrade to 1.1.19 and stop wasting time on 1.1.12.

[dhruvkej9]

Weird, my OpenCode is smart enough to clear drop cache of context to recover memory and continues to work. I am using desktop OpenCode. You should probably upgrade to 1.1.19 and stop wasting time on 1.1.12.

1.1.19 Have serious bugs when using GitHub copilot, it consumes multiple premium requests & has wiped out monthly credits of many people

[super-unique-ai]

Weird, my OpenCode is smart enough to clear drop cache of context to recover memory and continues to work. I am using desktop OpenCode. You should probably upgrade to 1.1.19 and stop wasting time on 1.1.12.

1.1.19 Have serious bugs when using GitHub copilot, it consumes multiple premium requests & has wiped out monthly credits of many people

Thanks for letting me know. 1.1.16 works well. 1.1.15 also works well.

[SteveyBoros]

OpenCode compaction needs serious work, even when it does work it's beyond terrible.
There are a bunch of good PR's on here but the powers that be apparently aren't interested.

Super weird.

[super-unique-ai]

OpenCode is still better than 99% of carp on GitHub.

[dhruvkej9]

OpenCode compaction needs serious work, even when it does work it's beyond terrible.
There are a bunch of good PR's on here but the powers that be apparently aren't interested.

Super weird.

@rekram1-node

[rekram1-node]

I can look into it, which PRs are you referring to?

It's weird cause some people say it's great, and I never hit it so I can imagine it leaves much to be desired haha

[super-unique-ai]

I can look into it, which PRs are you referring to?

It's weird cause some people say it's great, and I never hit it so I can imagine it leaves much to be desired haha

I think if they all try OpenCode with OpenCode Zen then they might realize it's their LLMs that are letting them down. I have $200 dollar a month ChatGPT account yet OpenCode Zen still does a better job at programming.

This very second OpenCode Zen MiniMax is beating ChatGPT Codex.

[SteveyBoros]

I can look into it, which PRs are you referring to?

It's weird cause some people say it's great, and I never hit it so I can imagine it leaves much to be desired haha

This is the most promising one: #4659

[dhruvkej9]

@rekram1-node Still the issue on 1.1.34

Here is the session's md file:

https://drive.google.com/file/d/1EBcJWzI5GR9pB-5bd7O5M3EJ6N2FJT7P/view?usp=sharing