I am using the latest available version of opencode-claude-bridge:
❯ cat ~/.cache/opencode/packages/opencode-claude-bridge@latest/package.json
{
"dependencies": {
"opencode-claude-bridge": "1.10.8"
}
}
I'm seeing some "unavailable tool" failures for tools that as per the README are currently handled by the shared-tool bridging feature:
% WebFetch https://github.com/Elmoaid/TempoGraph/tree/main/tempograph
⚙ invalid [tool=WebSearch, error=Model tried to call unavailable tool 'WebSearch'. Available tools: invalid, question, bash, read, glob, grep, edit, write, task, webfetch, todowrite, skill, compress, agent_attribution, envsitter_keys, envsitter_fingerprint, envsitter_match, envsitter_match_by_key, envsitter_scan, envsitter_validate, envsitter_copy, envsitter_format, envsitter_reorder, envsitter_annotate, envsitter_add, envsitter_set, envsitter_unset, envsitter_delete, envsitter_help, quota_status, websearch_cited.]
# I should note that WebFetch failures are inconsistent. In the same session, a previous invocation succeeded:
% WebFetch https://docs.devin.ai/work-with-devin/deepwiki-mcp
% WebFetch https://github.com/erajasekar/ai-diagram-maker-mcp
% WebFetch https://github.com/hashicorp/terraform-mcp-server
% WebFetch https://docs.ansible.com/projects/vscode-ansible/mcp/
% WebFetch https://github.com/danmartuszewski/hop
# It also seems like once WebFetch fails, the module does not try again
I'm also seeing a failure for the "question" tool though this is more confusing since the README mentions that "question" is mapped to "multiSelect/multiple" which I don't see listed in the output below:
⚙ invalid [tool=question, error=Model tried to call unavailable tool 'question'. Available tools: invalid, bash, read, glob, grep, edit, write, task, webfetch, todowrite, skill, compress, submit_plan, agent_attribution, envsitter_keys, envsitter_fingerprint, envsitter_match, envsitter_match_by_key, envsitter_scan, envsitter_validate, envsitter_copy, envsitter_format, envsitter_reorder, envsitter_annotate, envsitter_add, envsitter_set, envsitter_unset, envsitter_delete, envsitter_help, quota_status, websearch_cited, tempograph_prepare_context, tempograph_focus, tempograph_blast_radius, tempograph_diff_context, tempograph_symbols, tempograph_file_map, deepwiki_read_wiki_structure, deepwiki_read_wiki_contents, deepwiki_ask_question.]
FWIW, GPT 5.5 in OpenCode responds with a different tool name when prompted about being able to ask questions:
> Can you tell me what tools you have access to?
I have access to these tool groups:
- Shell/terminal: run commands with bash (git, npm, tests, etc.).
- Filesystem: read files/directories, glob file patterns, grep content, and apply patches.
- Planning/questions: maintain todos and ask you structured clarification questions.
- Subagents/skills:
- reviewer subagent
- skills such as refresh-docs, prompt-engineer, skill-creator, plugin-developer, etc.
- Web/docs:
- fetch URLs
- cited web search
- DeepWiki repository docs/questions
- Repo/code analysis:
- TempoGraph overview, focus, hotspots, dependencies, blast radius, diff context, symbol index, file map, dead code.
- Environment/.env-safe tools:
- list dotenv keys, validate syntax, compare/match values without printing secrets, add/set/delete/format keys.
- GitHub via CLI: I can use gh through the shell for PRs/issues/checks when needed.
- Context management: compress older conversation context.
- Diagnostics: quota/status diagnostics and agent attribution.
Build · GPT-5.5 · 7.9s
> What is the name of the tool for asking Questions?
The tool is named functions.question.
Version Info
- OpenCode 1.14.30
- WSL2 Version: Debian 13 (trixie) on WSL version: 2.6.3.0 / Kernel version: 6.6.87.2-1
- Windows Terminal version: Windows Terminal Preview / Version: 1.25.1171.0
- Windows Version: Windows 11 25H2 (26200.8246)
I am using the latest available version of opencode-claude-bridge:
I'm seeing some "unavailable tool" failures for tools that as per the README are currently handled by the shared-tool bridging feature:
I'm also seeing a failure for the "question" tool though this is more confusing since the README mentions that "question" is mapped to "multiSelect/multiple" which I don't see listed in the output below:
FWIW, GPT 5.5 in OpenCode responds with a different tool name when prompted about being able to ask questions:
Version Info