用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/tenstorrent/tt-forge --skill tt-bug-report命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | tt-bug-report |
| description | File a bug report with a reproducer against Tenstorrent repos (tt-lang, tt-metal, tt-xla) |
| argument-hint | <description of the problem> |
File a bug report for a Tenstorrent issue. Gather context, minimize the reproducer, and create a GitHub issue in the right repo.
| Component | Repo |
|---|---|
| TT-Lang (DSL, compiler, kernels) | tenstorrent/tt-lang |
| TTNN (ops, tensor APIs, device) | tenstorrent/tt-metal |
| TT-Metal (low-level C++ kernels) | tenstorrent/tt-metal |
| TT-XLA (PyTorch/JAX frontend) | tenstorrent/tt-xla |
If it's unclear which repo the bug belongs to, ask the user.
$ARGUMENTS
Determine the type of issue and gather relevant info:
For crashes or incorrect results:
For compiler errors (tt-lang):
/tmp/ttlang_initial.mlir, /tmp/ttlang_final.mlir)Ask the user if it's OK to upload the reproducer to the issue and spend some iterations trying to minimize it. If they agree, try to reduce it to the smallest script that still triggers the bug. Any reproducer is better than none -- if you can't minimize further, use what you have.
Use gh to create an issue in the appropriate repo:
gh issue create --repo <owner/repo> \
--title "[external bug report][from claude] <short title>" \
--body "$(cat <<'EOF'
<body>
EOF
)"
The body should include:
Return the issue URL to the user.