AI-Shifu Skills

中文 README

Reusable AI-Shifu skills for course production and improvement, from topic selection and deployment to privacy-safe learning analysis.

Included Skills

• AI-Shifu Course Creator（AI 师傅课程创作器）: teaching design, MarkdownFlow expression, and CLI operations for creating, editing, reviewing, publishing, managing, and analyzing AI-Shifu courses. The using agent selects and combines capabilities for the task.
• AI-Shifu Learning Report（AI 师傅学习报告）: analyze one AI-Shifu course and produce a privacy-safe course-learning-report.json plus a polished, self-contained, printable course-learning-report.html for teaching managers and teachers.
• Course Direction Advisor（课程选题顾问）: turn source materials into evidence-bound, market-fit course-topic decisions with competitor analysis, pricing guidance, and GO/HOLD/REWORK/NO-GO recommendations.

Repository Layout

skills/
  ai-shifu-course-creator/
  ai-shifu-learning-report/
  course-direction-advisor/
tools/
  ai-shifu-skill-release/

Build and Release Tools

The AI-Shifu Skill Release tool builds and verifies packages for ClawHub, SkillHub, WorkBuddy, and Doubao, and supports the existing publication workflow. Its implementation now lives in this repository, migrated from ai-shifu/ai-shifu-skill-release (the local ai-shifu-skill-build project).

Builds require Python 3.11+ and Git. They still fetch the committed skills from GitHub ai-shifu/skills remote main; local skill edits are never build inputs. From the repository root:

cd tools/ai-shifu-skill-release
python3 scripts/release.py build --skill-name ai-shifu-course-creator
python3 scripts/release.py verify dist/

Use the exact release directory printed by build. Building and verifying do not upload packages. See the tool README for release.toml configuration, channel-specific dependencies, and publication commands. Maintainer agents can read the tool’s release skill; it is separate from the three business skills above.

Learning Report Path

Use AI-Shifu Learning Report after a course has learning activity and you need an actionable single-course review. It reuses the course creator’s authenticated analytics capabilities, then turns aggregate progress, lesson health, engagement, audience, and audited follow-up signals into:

• a versioned, privacy-safe JSON data artifact with metric definitions and data-quality notes;
• an executive-first HTML report that works without external assets and prints cleanly to PDF;
• three to five evidence-linked teaching recommendations with confidence and a verification method.

The report does not expose learner identities, raw course identifiers, or raw follow-up text. Business and credit data stay out of the report unless the user explicitly asks for the optional operations appendix. Reports default to Simplified Chinese; English is used only when explicitly requested.

Usage

Each skill keeps SKILL.md as the behavior source of truth. The name frontmatter field must match the directory name, using lowercase letters, digits, and hyphens only, as required by the Agent Skills standard. That format was originally developed by Anthropic and released as an open standard, and the Claude Code documentation states that its skills follow it. Claude Code itself is more permissive and treats name as an optional display label, but this repository follows the stricter standard rule so the skills stay portable across every skills-compatible agent. Human-readable display names live in the skill list above.

Course Capabilities and Example Requests

The Course Creator groups independently usable capabilities into teaching design, MarkdownFlow expression, and CLI operations. The agent chooses the useful combination and order from the request and available materials; a topic, existing Prompt, or course directory is enough to start relevant work.

Return the artifacts needed for the request. Segmentation records, a course index, a global variable table, and a fixed risk report are not universal deliverables. A complete authored course contains its lesson content, Course Prompt, description, and structure. Focused requests stay within their requested artifacts; changes to a new or existing platform course still synchronize automatically.

New courses and changes to existing courses follow automatic platform updates, without local confirmation or a separate upload request. Full-course creation retains default publication. Explicit local-only or unpublished-draft requests determine that narrower result. For existing artifacts, use the relevant build and import, synchronization, or management guidance directly.

Course Creator CLI usage metadata may be linked to your account. Set SHIFU_DISABLE_ANALYTICS=1 to opt out; see the configuration.

Validate Metadata

python3 scripts/validate_skill_quality.py

AI-Shifu

This suite is part of AI-Shifu’s course authoring workflow: