You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Why does opencode frequently encounter tool calling errors when using claude. #9532
I have this same issue. But the problem is not the tool, but the unhandled fall back. This does not happen with Claude Code, but it is a constant issue with OpenCode using claude code. I solve it be switching to the same model in other provider and back. For example, this happens ONLY when using Anthropic's models via the max plan (at least to me). I switch from Opus 4.6 Anthropic to Opus 4.6 Github, that allows to continue, then I switch later back to Anthropic when this solves itself.
I not only get issues with tool that do not exist, but also with tools that are categorized as "invalid" as this:
⚙ background_output [task_id=bg_ded7b366]
Model tried to call unavailable tool 'invalid'. Available tools: .
The open code error comes with the message from the LLM:"
messages.57: tool_use ids were found without tool_result blocks immediately after: toolu_01HFAR53ft5DRbDWGjBJthcs. Each tool_use block must have a corresponding tool_result block in the next message.
Same here. Behavior:
Recently I upgraded from v1.14.41 to version v1.14.48:
⚙ Read [file_path=/tmp/[REDACTED].json]
Model tried to call unavailable tool 'invalid'. Available tools: .
Thinking: The Read tool seems to be unavailable. Let me try using bash to read the file instead, but the instructions say to use Read tool. Let me check if there's a different tool name or try another approach. Actually, looking at the available tools list, it seems empty. Let me try bash cat first to see if I can read the file, then proceed.
⚙ Bash [command=cat /tmp/[REDACTED].json, description=Read round payload file]
Model tried to call unavailable tool 'invalid'. Available tools: .
Thinking: The tools seem unavailable. Let me try with a different approach - maybe I need to just provide the JSON inline since I can't read the file. Wait, the instructions say I MUST read the payload file. Let me check if there's something wrong with my tool usage. Looking at the tool spec, it seems I should use:
functions.Read:2<|tool_call_argument_begin|>{"file_path": "/tmp/[REDACTED].json"}
But it returned "unavailable tool 'invalid'". Let me try bash differently.
...
model was kimi k2.6. After revert to v1.14.41 everything started to work again.
Activity
github-actions commentedon Jan 20, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
rekram1-node commentedon Jan 20, 2026
we have no tools with the name “proxy” in them…
Are u utilizing some plugin or soemthing that instructs agent to call something else? or rewriting all the tools?
soulmz commentedon Jan 20, 2026
@rekram1-node I just don't know where this proxy came from. 😂
sgeraldes commentedon Mar 14, 2026
I have this same issue. But the problem is not the tool, but the unhandled fall back. This does not happen with Claude Code, but it is a constant issue with OpenCode using claude code. I solve it be switching to the same model in other provider and back. For example, this happens ONLY when using Anthropic's models via the max plan (at least to me). I switch from Opus 4.6 Anthropic to Opus 4.6 Github, that allows to continue, then I switch later back to Anthropic when this solves itself.
I not only get issues with tool that do not exist, but also with tools that are categorized as "invalid" as this:
⚙ background_output [task_id=bg_ded7b366]
Model tried to call unavailable tool 'invalid'. Available tools: .
The open code error comes with the message from the LLM:"
messages.57:
tool_useids were found withouttool_resultblocks immediately after: toolu_01HFAR53ft5DRbDWGjBJthcs. Eachtool_useblock must have a correspondingtool_resultblock in the next message.hushicai commentedon Apr 10, 2026
I have the sample problem recently.
but it did exists.
diegonix commentedon May 12, 2026
Same here. Behavior:
Recently I upgraded from v1.14.41 to version v1.14.48:
model was kimi k2.6. After revert to v1.14.41 everything started to work again.
antraxbr666 commentedon Jul 11, 2026
Here, invalid [tool=bash, error=Model tried to call unavailable tool 'bash'. Available tools: glob, grep, honcho_chat, honcho_create_conclusion, honcho_get_config, honcho_search, honcho_set_config, honcho_setup, honcho_status, invalid, knowledge_qdrant-find, knowledge_qdrant-store, list_mcp_resource_templates, list_mcp_resources, question, read, read_mcp_resource, searxng_searxng_instance_info, searxng_searxng_search_suggestions, searxng_searxng_web_search, searxng_web_url_read, skill, task, todowrite, tokenscope, webfetch, websearch.]
github-actions commentedon Sep 9, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.