OpenGHz

Rebase a local branch onto upstream/main and handle conflicts safely. Use when user says "upstream 更新了", "git pull --rebase upstream main", "rebase upstream/main", "同步上游 main", "处理 rebase 冲突", or wants to update a fork from upstream with automatic conflict resolution.

Configure Git remotes for a fork workflow. Use when user says "add my fork", "set origin upstream", "fork remote", "git remote", "upstream remote", "像 origin/upstream 这样", or wants origin to point to their fork and upstream to the original repository.

Produce a structured record of an agent's command-execution process — the commands to be run, their purpose, results summaries, and full terminal output saved to log files. Triggers on explicit user requests like "过程记录", "做个过程记录", "记录这次执行过程", "process record", "document this run", "trace this workflow", or similar. Always use this skill when the user invokes it by name or asks for a recorded/documented run, even if the underlying task seems trivial. Plans are written before execution, the agent pauses for confirmation on risky/expensive steps, and results are filled in as commands run.

This skill should be used when the user asks to "update documentation for my changes", "check docs for this PR", "what docs need updating", "sync docs with code", "scaffold docs for this feature", "document this feature", "review docs completeness", "add docs for this change", "what documentation is affected", "docs impact", or mentions documentation updates in any project. Provides a guided workflow for updating project documentation based on code changes.

Comprehensive package and environment management using pixi - a fast, modern, cross-platform package manager. Use when working with pixi projects for (1) Project initialization and configuration, (2) Package management (adding, removing, updating conda/PyPI packages), (3) Environment management (creating, activating, managing multiple environments), (4) Feature management (defining and composing feature sets), (5) Task execution and management, (6) Global tool installation, (7) Dependency resolution and lock file management, or any other pixi-related operations. Supports Python, C++, R, Rust, Node.js and other languages via conda-forge ecosystem.

Professional embodied-AI paper-writing coach distilled from 63 top-conference papers (CoRL, RSS, ICRA, IROS, Science Robotics, 2022–2026). Teaches writing craft only — vocabulary, sentence patterns, paragraph flow, figure/table conventions, section-by-section construction, rhetorical pivots, appendix norms. NOT a content advisor: teaches HOW to write, not WHAT to claim. Use when the user mentions writing or reviewing any paper section (abstract, intro, method, related work, experiments, results, ablations, conclusion, limitations, appendix), titling, figure captioning, paragraph polishing, rebuttals, or submission prep. English cues: "write my abstract", "draft an intro", "title my paper", "caption this figure", "review this section", "fix my method", "polish this paragraph", "limitations section", "help with my rebuttal", "ready for submission". Chinese cues: 「帮我写摘要」「润色引言」「修改这段」「figure怎么标caption」「method怎么组织」「experiments怎么写」「conclusion怎么收尾」「rebuttal怎么写」「投稿前帮我看一下」「像不像顶会风格」.

Turn a project folder into a polished open source project. Use when: open source ready, polish README, default MIT license, add LICENSE/CONTRIBUTING/CODE_OF_CONDUCT/CHANGELOG, automatically create public GitHub remote for projects without git remote, prepare repository for public release. Only add or modify project metadata/docs, never change source code.

Remove image backgrounds with rembg, producing transparent-PNG cutouts. Use this whenever the user wants to remove/erase/delete a background, cut out or isolate a subject, make a transparent PNG, extract a foreground object, or batch-process a folder of images into cutouts — including phrasings like "抠图", "去背景", "remove background", "cut out this object", "make this transparent", "isolate the product", or when they hit rembg errors (missing onnxruntime backend, CUDA/libcudart version mismatch, gradio proxy crash). It sets up an ISOLATED conda/venv environment (never touching existing envs), installs the correct GPU (CUDA 12) or CPU backend for the machine, prefetches models, and runs single-image or whole-folder background removal.

Author and run resumable, cached install/setup scripts with the `anchored-install` (`ai`) tool — a backend-agnostic installer that gives plain shell scripts Dockerfile-style layer caching, so a failed re-run resumes from the last successful step instead of starting over from the top. Use this whenever the user wants an install, setup, bootstrap, provisioning, or environment-setup script that is resumable, restartable, idempotent, re-runnable, or cached — or that "shouldn't redo everything that already worked" when it fails and is re-run — even if they never mention anchored-install or `ai` by name. Also use it when writing or editing `#ai` manifests, debugging a manifest that failed mid-run, or converting an existing install.sh / setup script into an anchored-install manifest.

Install agent skills with `npx skills` the safe, no-duplicate-work way. Before installing, it checks (in one shot) whether the skill is already installed and whether every agent's skills directory is symlinked onto one shared hub — so you install ONCE instead of once per agent. Use this whenever the user wants to add, install, or set up a skill via `npx skills` / the skills CLI, hands over a skills repo or a skill name to install, asks why a skill isn't showing up in some agents, or wants to share one skills folder across Claude Code / Codex / Cursor / Gemini / Copilot / OpenCode. Triggers on: "install this skill", "add the X skill", "npx skills add", "set up this skills repo", "安装这个技能", "装一下这个 skill", "把这个技能装上", "用 npx skills 安装", "让所有 agent 都能用这个技能", "技能装了吗 / 是不是已经安装了", "把各个 agent 的 skills 目录软链接到一起".

Record friction and bugs encountered while using or operating any project, into an append-only Markdown log (AGENT_ISSUES.md) that the user can later send back to the project's author for fixes. Trigger ONLY when the user explicitly asks for it — either by invoking this skill by name, or by clearly telling you to record/collect usage problems, e.g. "记录这个问题", "记一下这个 bug", "把踩的坑记下来", "接下来把用这个项目遇到的问题都记录下来", "log that bug", "note this for the author", "write down what went wrong", "from now on log any issues you hit while using this repo", or "collect the issues you ran into testing this tool". Do NOT trigger on your own: if the user has not asked you to record issues, do not start a log just because you happened to hit a bug — handle the problem as part of the normal task instead. Once the user has explicitly turned issue-recording on, keep logging qualifying problems you encounter for the rest of that session. This skill is project-agnostic — it works for any codebase, CLI, or tool you're operating.

Write Python configuration, command-line, and class-initialization code in the user's house style: pydantic config classes (frozen, with attribute docstrings as the field docs), Hydra for whole-application config wiring via `_target_` + `instantiate`, pydantic-settings (`CliApp`) for small-script CLIs, and a single `config` parameter in every `__init__`. Use this whenever writing or refactoring Python that defines a configuration model, parses command-line arguments, structures a class or top-level function around its settings, or chooses between Hydra and pydantic-settings — even if the user never names the libraries. Triggers on: "add a config", "write a config class", "make a CLI for this script", "add command-line args", "pydantic config", "hydra config", "refactor this to take a config", "配置类", "加个配置", "写个命令行", "加命令行参数", "把它改成吃 config 的", and similar Python config / CLI / class-init work.