Harzva

Elsevier Ablation Study: ablation design, component-level evidence, sensitivity analysis, and mechanism validation.

Elsevier Abstract + Keywords: abstract, keywords, contribution compression, and discoverability for Elsevier submissions.

Elsevier Abstract: abstract drafting, compression, claim discipline, and contribution evidence alignment.

Elsevier Algorithm Pseudocode: algorithm boxes, pseudocode, complexity notes, and stepwise method communication.

Elsevier Appendix: appendices, extended derivations, extra tables, and material that should not interrupt the main paper.

Elsevier Baseline Selection: baseline choice, reimplementation fairness, hyperparameter disclosure, and comparison scope.

Elsevier Caption: figure and table captions, legends, abbreviation expansion, symbol explanation, and standalone interpretability.

Elsevier Checklist Risk Register: pre-submission checklist, reviewer-risk register, desk-rejection risk, and issue prioritization.

Use when a project needs a standard .claude workspace initialized or repaired incrementally. This skill audits the existing .claude folder, adds missing files and directories without overwriting existing rules/skills/settings, and can optionally evolve the structure over time for the current project.

Generate Zhihu article in the style of 22-庖丁解牛专题页-知乎图文终版.md

Use when the user wants to search GitHub repositories by keyword, inspect ranked candidates, compare likely matches, and optionally clone the selected repository into a specified local directory. Good for prompts like “帮我搜 GitHub 上相关仓库”, “帮我找并下载一个相关仓库”, “搜索 GitHub 上的项目然后拉到 reference”, or “按关键词找仓库并 clone 到指定位置”.

Use when the task is to build or maintain a repo-local 热门话题/AI日报专项, aggregate multiple public hot-topic sources such as AIbase Daily or RSS feeds, normalize them into a reusable snapshot, route the findings into the best existing site hotspot pages, and keep the hot-topic hub updated over time.

Use when the task is to take a user-provided keyword, search the web for relevant articles, summarize and cluster the findings, choose the most appropriate existing `site/` topic or hotspot page in this project, and draft a publish-ready Markdown article with explicit image prompt placeholders or Mermaid notes.

Use when a task should advance through repeated plan-driven loop cycles for an engineering project, paper, or long-running repo change. loloop wraps the official /loop idea with the repo's loop-in-loop method: read the current plan, inspect prior evolution notes, execute the current iteration, write a new evolution record, draft the next plan when a version is complete, and prepare the next /loop handoff.

Automatically fix, enhance, and optimize other skills when problems occur. Use when a skill fails or produces unexpected results. Prevents similar issues in the future.

Diagnose a Zhihu Markdown draft before publishing. Use when a Zhihu article needs a final copy audit for AI-flavored phrasing, professionalism, readability, and evidence strength, especially before calling zhihu-publish.

Validate Android release APKs with a local emulator and adb: detect MuMu/Android emulator, install APKs, launch apps, capture screenshots, UI XML, logcat, SHA256, and release-artifact evidence, then report QA readiness. Use when asked to test APKs, verify Android release builds, diagnose emulator install/launch/UI issues, check whether APK artifacts match a GitHub Release, or package a repeatable Android QA workflow. Recommended Windows emulator download: https://mumu.163.com/360mumu/.

Create local, browser-openable preview labs for packaged apps, including APK WebView previews, IPA or iOS concepts, desktop installer previews, and mobile UI state mirrors. Use when asked to avoid repeated package installation, preview app UI/content in HTML, compare Android/iOS/desktop shells, add theme or language toggles, or document release artifact previews.

Create polished mobile app UI concepts, high-fidelity single-screen prototypes, multi-screen app theme suites, design systems, screenshots, and matching source code for iOS, Android, and mobile web apps. Use when Codex is asked for app UI design, mobile screen concepts, app design case studies, UI inspiration research, README-ready screenshot galleries, source-linked UI catalogs, or HTML/CSS/React source for mobile app screens.

Install, select, adapt, and apply DESIGN.md design-system files as a repeatable AI frontend workflow. Use when the user mentions DESIGN.md, getdesign.md, awesome-design-md, Google Stitch DESIGN.md, brand-inspired UI generation, installing a design style into a project, or turning a design reference into a workflow skill for agents.

Evidence-first Windows workbench smoothness and token-saver workflow. Use when the user asks to make a Windows workspace, local AI/product folder, release workspace, or desktop packaging loop smoother by reducing duplicate artifacts, missing release proof, encoding failures, repeated context loading, and project re-entry cost.

Design and rewrite polished GitHub README homepages for repositories and profile READMEs, especially Chinese/English open-source projects. Use when the user asks to beautify, redesign, improve, package, market, add a README logo or wordmark, or make a README more professional, visual, persuasive, scannable, or GitHub Pages-like. Focus on README aesthetics, layout, badges, screenshots, logo/wordmark identity, project positioning, information architecture, and Markdown/HTML that renders well on GitHub.

Generate polished README-ready screenshots, GIFs, videos, hero composites, and interactive HTML preview galleries for local web, mobile preview, desktop preview, packaged-app preview, and documentation projects. Use when Codex needs to update README visuals, capture UI states, record product demos, standardize docs screenshots, build preview galleries, or make a GitHub project presentation stronger.

Use when promoting GitHub repositories on README and Xiaohongshu with a consistent bright infographic style: horizontal README banner, vertical Xiaohongshu carousel image, public-safe copywriting, MCP publishing, and local publication records.

Use when planning, creating, reviewing, publishing, or recording Xiaohongshu notes for GitHub repositories, AI tools, AgentOS stacks, MCP tools, and open-source developer projects. Routes to promo-card generation, content quality review, MCP publishing workflow, and publication ledger skills.

Use when running an end-to-end Xiaohongshu campaign for GitHub repositories, AI tools, AgentOS stacks, MCP tools, or open-source projects from repo facts through images, copy review, optional MCP publishing, and local publication records.

Use before publishing Xiaohongshu notes to improve title, opening hook, body structure, hashtags, cover text, evidence quality, and compliance for GitHub repositories, AI tools, AgentOS stacks, MCP tools, and developer products.

Use when recording Xiaohongshu drafts, publish attempts, successful notes, MCP results, image assets, repository links, and sanitized publication evidence in a local ledger.

Use when publishing a Xiaohongshu note through the local Xiaohongshu MCP, checking login status, preparing MCP publish arguments, or recording sanitized publish results. Does not store credentials or login state.

Bring web-login, CLI-first AI capabilities back into a Claude Code or LikeCode-centered workflow. Use when you want to call Codex, Gemini, Qwen, or Grok as external workers from within Claude Code.

Use Codex as an external CLI worker inside a Claude Code or LikeCode-centered workflow. Good for implementation-oriented feedback, code-aware review, and execution-heavy coding tasks.

Use Gemini as an external CLI worker inside a Claude Code or LikeCode-centered workflow. Good for broad review, first-pass critique, and structured feedback.

Use Grok as an external CLI worker inside a Claude Code or LikeCode-centered workflow. Good for critical review and alternative perspectives. Requires GROK_API_KEY or equivalent local configuration.

Use Qwen as an external CLI worker inside a Claude Code or LikeCode-centered workflow. Good for secondary critique, math reasoning, and multilingual tasks. May require device authorization on first run.

Create public-safe GitHub README promo banners and Xiaohongshu vertical cards for academic rebuttal skill suites. Use when promoting rebuttal-skill-suite, turning GitHub/README/web screenshots into Image2-style marketing graphics, preparing Xiaohongshu copy, or logging Xiaohongshu publication evidence.

Build private and sanitized dashboard data for Rebuttal Skill Suite by scanning a project, paper, reviews, rebuttal drafts, evidence ledgers, gate outputs, and persona feedback, then producing a JSON snapshot that the static webview can load.

Audit and polish one-page academic rebuttals or author responses, especially LaTeX/PDF responses with reviewer IDs, strict page limits, compact evidence tables, reviewer-specific concerns, protocol or label-use ambiguity, cost evidence, negative-case framing, and strict short-tail-line/orphan-word checks. Use when Codex must check formatting, page fullness, orphan lines, reviewer coverage, cited evidence, rebuttal tone, or final submission readiness.

Audit academic rebuttals, author responses, LaTeX/PDF text, Markdown drafts, and OpenReview replies for leaked internal/developer notes, local paths, tool traces, private experiment logistics, advisor or AI feedback traces, non-reviewer-facing remarks, and wording that should be rewritten before showing reviewers or ACs. Use when checking whether rebuttal content accidentally exposes Codex/ChatGPT/Claude notes, teacher/advisor comments, TODOs, logs, shell commands, filesystem paths, “AC-facing” meta-language, draft-only reasoning, or score-strategy language.

Plan-driven recurring iteration for Codex without a native /loop command. Use when Codex should keep advancing the same project across repeated passes, read an active plan, inspect prior evolution notes, execute one bounded slice, write the next evolution note, prepare the next handoff, or run a periodic automation daemon that resumes the same Codex thread.

Reflective supervisor and worker workflow for Codex-Managed-Agent team spaces. Use when working with .codex-team, assigning or reviewing mailbox tasks, checking blocked/stale/review work, maintaining failure_bank.json, creating bounded worker prompts, or preparing the next supervisor handoff.

Orchestrate Harzva app work across product UI, mobile and desktop implementation, local preview, emulator QA, release packaging, online update manifests, latest-download buttons, and safe overwrite-install behavior. Use when creating, redesigning, testing, releasing, or standardizing any app or app-like product.

Discover trustworthy GitHub, Gitee, and GitCode repositories, Releases, installable assets, Codex skills, and MCP servers for AI-agent tasks. Use when an agent needs to find upstream software, compare release assets, inspect license/safety signals, locate skills/MCP integrations, or choose what to install or cite.

Delegate small, bounded, non-multimodal implementation, review, README, docs, prompt, checklist, and mechanical-edit subtasks to codex-5.3-spark / gpt-5.3-codex-spark while keeping planning, final review, release decisions, publishing, device QA, visual inspection, and dangerous operations in the parent Codex session.

Codex Spark shortcut skill for delegating small, bounded, non-multimodal coding, docs, README, prompt, checklist, review, or mechanical-edit tasks to a Spark sub-agent using gpt-5.3-codex-spark. Use when the user says cxspark, cx spark, Codex Spark shortcut, or asks to run a narrow task through Spark while the parent Codex session keeps planning, review, verification, publishing, and dangerous operations.

Use when writing, normalizing, splitting, auditing, or updating roadmp/roadmap Markdown files in the Just-DDL roadmpxx.md style: structured milestones, task files, checkbox governance, evidence-based completion, staged audits, candidate tables, assumptions, blockers, and Chinese production-roadmap prose.

Scan and package-manage a local AgentWorkOS environment, including Codex skills, agent role cards, TERMS.md, AGENTS.md rules, local repositories, lockfiles, and sync drift. Use when the user asks to migrate machines, restore AI workspace context, package skills, or verify three-end sync.

Inventory locally configured GitHub accounts and repositories, map online GitHub repositories to local Git project folders, verify whether local branches match fetched upstream commits, batch-fetch GitHub Pages and release metadata, and produce Markdown/JSON reports for README optimization, repository cleanup, publishing audits, Pages matrices, release hubs, or account migration work. Use when the user asks to list all GitHub repos, find local project locations for GitHub repos, check local-vs-remote version status, audit README targets, collect Pages URLs/releases, or organize repos across Harzva/saihao/Just-Agent style managed accounts.

Generate and query GitHub remote-to-local repository maps without requiring the RepoAtlas desktop app. Bundles gh-repo-cartographer for repository inventory, local checkout matching, compact maps, address-cache refreshes, and optional MCP lookup.

Sets up and manages an autonomous scheduler loop based on reflective machine learning metaphors (Pretrained Backbone, Local Adapters, Error Banks). Use when asked to setup or manage a gemini-loop.