Tool execution frequently fails with 'Tool execution aborted' error

Bug Description

Tools (especially bash, edit, read) frequently fail with 'Tool execution aborted' error in opencode v1.3.0.

Steps to Reproduce

Run any bash command
After a few successful calls, the tool starts returning "Tool execution aborted" error
Must wait a while or restart session to recover

Expected Behavior

Tools should execute reliably without random aborts

Environment

OS: Linux (Ubuntu 20.04)
opencode version: 1.3.0
Model: minimax-m2.5-free

Additional Info

This happens across multiple tool types - bash, edit, read, grep all experience intermittent failures. The error occurs without any error message and completely blocks tool usage.

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

tool execution aborted #17159: Same title 'tool execution aborted' - intermittent tool execution failures
[BUG] Write tool continously getting aborted #11079: Write tool continuously getting aborted with similar symptoms across tool types
Tool execution aborted when tool call is truncated (empty input) #13102: Tool execution aborted when tool call is truncated (empty input)
Bug: Session corrupted when tool execution is aborted - missing tool_result causes API rejection #8312: Session corrupted when tool execution is aborted (closed)

Please review these issues to see if your problem is already tracked, and consider adding a comment there instead if it matches your situation.

This is happening to me also.

OS: MacOS Sequoia
opencode version: 1.2.27
Model: Opus 4.6

I am experiencing the same on Ubuntu 26.04
Model GPT 5.4 Mini - Medium

can anyone send sessions they encountered this in? opencode export > session.json

@rekram1-node

@rekram1-node I have a session export where I have run into this a handful of times. Where should I send it to?

This is now happening to me in every chat I have on opencode. I don't have a session I'm able to share due to the work I'm doing, but after recent updates (1.16.x, 1.17.x) it happens whether i'm on qwen3.6-35b-a3b (my favorite model) or even kimi k2.6. Very strange. Both worked very well before but now more often than not, I get tool call aborted, unknown tool called, etc etc.

I suspected maybe this was an issue with specific opencode agents i'm using - but it happens on the default Build/Plan agents / modes as well.

@rekram1-node

Adding a bit more detail in case it helps narrow this down.

From what I’ve seen, this issue seems to start happening after the chat/session gets fairly lengthy, usually after at least one compaction.

The first sign is not always a tool failure right away. Sometimes the response gets interrupted or does not finish cleanly. It can cut off midstream, so I’ll ask it to continue or finish the response, and then a message or two later the tool calls may start erroring out or failing.

My setup is a pretty plain/fresh install of opencode, and I’ve mainly been using MiniMax M3. In my case, it usually seems to start showing up around ~$2 worth of usage on that model/session.

Not sure if this is the exact same circumstance others are running into, but I figured it was worth adding since it may point to longer sessions/compaction being part of the trigger.

@rekram1-node I still have a session export where I ran into this a handful of times. Happy to send it wherever is most useful.

I'm seeing this same "Tool execution aborted" error consistently, with a trigger pattern that might help narrow it down: every abort in my session was the second of two tool calls issued in quick succession, regardless of tool type or transport:

Two MCP tool calls to a remote HTTP server (Todoist MCP) fired together — second one aborted
Two MCP tool calls to a local stdio server (brave-search) fired close together — second one aborted
Two calls to a native tool with no MCP involved at all (webfetch) — second one aborted

This happened consistently (7/7 times observed) across completely different tool types and transports, which suggests it's not a per-server session/transport issue but something in the tool-dispatch/scheduler layer canceling a call that starts while another is still resolving.

Environment:

opencode version: 1.18.2
OS: macOS (Darwin 25.5.0)
Reproduces with both parallel (same-message) and rapid back-to-back sequential tool calls

Workaround: spacing out tool calls (avoiding near-simultaneous dispatch) avoids the issue.

I am also hitting this issue, same pattern as above -- simply asking the agent to "space out tool calls so that they are not dispatched near-simultaneously" actually seemed to help. This was on anthropic models, seemed to work fine with deepseek ones.

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

Navigation Menu

Repository navigation

Tool execution frequently fails with 'Tool execution aborted' error #18757

Bug Description

Steps to Reproduce

Expected Behavior

Environment

Additional Info

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Tool execution frequently fails with 'Tool execution aborted' error #18757

Description

Bug Description

Steps to Reproduce

Expected Behavior

Environment

Additional Info

Activity

github-actions commented on Mar 23, 2026

jonathantullett commented on Mar 23, 2026

khi4 commented on May 30, 2026

rekram1-node commented on May 30, 2026

CorbinStewart commented on Jun 9, 2026

drobin04 commented on Jun 11, 2026

CorbinStewart commented on Jun 13, 2026

robertn702 commented on Jul 16, 2026

xnhp commented on Jul 17, 2026

github-actions commented on Sep 17, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions