菜单
Researches about a new method (scalarizer or aggregator) from the scientific literature and how to integrate it to TorchJD. Use when a contributor is interested into adding a new aggregator, scalarizer, or a new method that could be either, that is not already listed in the tracking issues.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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/SimplexLab/TorchJD --skill research-method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。
Implements a new method (scalarizer or aggregator) in TorchJD, starting from the research produced by the research-method skill and following the established file-by-file conventions. Use when a contributor wants to add the actual implementation of a scalarizer or aggregator that has already been investigated and listed in the tracking issues.
Verifies that a TorchJD release was published correctly by checking the docs site, installing from PyPI, and smoke-testing newly added classes. Use after a release has been merged and published.
Prepares a torchjd package release by verifying the changelog, README, bumping the version in pyproject.toml, and opening a release PR. Use when a maintainer asks to prepare a release or to release.
基于 SOC 职业分类
正在显示 SKILL.md
| name | research-method |
| description | Researches about a new method (scalarizer or aggregator) from the scientific literature and how to integrate it to TorchJD. Use when a contributor is interested into adding a new aggregator, scalarizer, or a new method that could be either, that is not already listed in the tracking issues. |
This skill researches about a new method by reading the paper and the existing implementations, proposing a plan for integrating an implmentation in TorchJD.
For agents: invoke as /implement-method method-name (paper-name) (e.g. /implement-method upgrad (Jacobian Descent for Multi-objective Optimization)).
If no method name is provided, ask the user for the name of the method and the title of the paper.
For humans: follow the numbered steps below to guide you in your development.
On GitHub, we have a tracking issue for scalarizers (https://github.com/SimplexLab/TorchJD/issues/667) and one for aggregators (https://github.com/SimplexLab/TorchJD/issues/665). The first goal is to generate a new row for the table in the appropriate tracking issue. You will also have to find the official implementation (if any), that could be linked in the paper, and the most known non-official implementations (search in particular in LibMTL, libmoon and pymoo, and search for more repos online). You will have to find the exact files and lines in which the method is implemented. You can also write special remarks if anything deserves further attention / if you're not sure about something. Do not bother trying to find citation count or the exact venue in which the paper was published, unless you directly see this information from your initial search.
Based on the paper and the existing implementations, find the things in the interface of the new method that are non-standard compared to existing methods. For example, is the method stateful (and in which way), random, does it require some warm-up period, other stats than those provided through the forward method (like losses for aggregators), and so on. Report all of those non-standard things.
Produce and output a row that can be directly copy-pasted to this table, but do not modify the issue yourself. Propose to post a comment on the issue with the row.