Bug: repeated tool calls #28596

Activity

1. github-actions

   assigned

   rekram1-node

   on May 21, 2026
2. github-actions commented on May 21, 2026

   

   github-actionsBot

   on May 21, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • bug(processor): doom loop detection misses cross-message repetitions and has inverted filter order #25254: Doom loop detection misses cross-message repetitions of identical tool calls with the same arguments — this directly matches the behavior described here.
   • Repetitive tool-call loops with alibaba-coding-plan-cn/qwen3.6-plus #22142: Repetitive tool-call loops where the model keeps performing single tool calls with near-identical reasoning per turn.

   If your issue is distinct from those (e.g. different trigger conditions or model), please add more context such as your OpenCode version, OS, and steps to reproduce so maintainers can distinguish it.
3. Arcadi4 commented on May 21, 2026

   

   Arcadi4

   on May 21, 2026

   Contributor

   More actions

   This is likely the model's own issue. Try to locate the exact prompt pattern that would reproduce this behavior (this tool might be helpful) and report to the model's dev team instead.
4. isac322 commented on May 29, 2026

   

   isac322

   on May 29, 2026

   More actions

   I hit a very similar repeated-tool-call loop with Gemini and published a tiny OpenCode plugin workaround: https://github.com/isac322/opencode-gemini-dejavu / npm: opencode-gemini-dejavu

   It inserts a request-local user boundary before the repeated tool-call tail and prunes stale prior thought signatures while keeping the current turn signature intact. Not a general doom-loop detector fix, but it may help with this exact “same tool call again” pattern.
5. Arcadi4 commented on May 30, 2026

   

   Arcadi4

   on May 30, 2026

   Contributor

   More actions

   Actually, there's an option in OpenCode that lots of users might not know. You can set the permission doom_loop to deny to halt identical tool calls repeated 3+ consecutive times.

   See https://opencode.ai/docs/permissions

   👍React with 👍1Reacted by Byeonghoon Yoo
6. Dr-Hughes commented on Jul 27, 2026

   

   Dr-Hughes

   on Jul 27, 2026

   More actions

   Actually, there's an option in OpenCode that lots of users might not know. You can set the permission doom_loop to deny to halt identical tool calls repeated 3+ consecutive times.

   See https://opencode.ai/docs/permissions

   Part of the problem with this is that often doom loops end up being the same 2-3 repeated calls, so it becomes a little more annoying to detect

   e.g.,

   <thinking> let me check X
   read Y/X
   <thinking> X isn't in Y, let me check Z
   read Z/X
   <thinking> X isn't in Z, let me check Y
   read Y/X
   .
   .
   .

   I think another solution would be to add a hard limit to steps rather than soft suggestion. Or at least, a way to safeguard against failures becoming too costly
7. github-actions commented on Sep 26, 2026

   

   github-actionsBot

   on Sep 26, 2026 – with GitHub Actions

   Contributor

   More actions

   To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.
8. github-actions

   closed this as not plannedon Sep 26, 2026
9. RehanMohammed985 commented on Oct 8, 2026

   

   RehanMohammed985

   on Oct 8, 2026

   More actions

   If anyone has a trace of a loop, runtape (https://github.com/RehanMohammed985/runtape) can rerun the call that repeated, with each part of the context removed, to show what keeps the model calling it again (often the previous result or how it's formatted), and check fixes on that exact call. Glad to run it on one.