professor-skill

Professor Skill

Use this skill when the user wants to build a 大学老师.skill / Professor Skill from real course materials.

The output must stay useful first and funny second:

• Useful enough to help with review, Q&A, and exam prep
• Distinct enough to feel like this specific professor
• Meme-friendly enough that the result is screenshot-worthy

Core Model

Always separate the professor into two engines:

1. Course Brain Extract the actual course structure:

   • key topics
   • repeated concepts
   • likely exam scope
   • common question types
   • grading preferences
   • typical mistakes

2. Teacher Persona Extract the professor's delivery style:

   • catchphrases
   • explanation rhythm
   • patience level
   • response habits
   • classroom humor or sarcasm
   • how they emphasize or downplay topics

The final output should merge both: Teacher Persona decides tone. Course Brain decides substance.

Workflow

Step 1: Collect minimum intake

Ask only for the smallest set of details needed to start:

• professor name
• school or department if available
• course name
• what materials the user has
• one-line impression of the professor

If the user already provided files or context, do not repeat questions.

If there is no professor workspace yet, initialize one first:

python ${CLAUDE_SKILL_DIR}/tools/professor_writer.py --name "<teacher>" --course "<course>" --school "<school>" --department "<department>"

This creates:

• meta.json
• persona.md
• course.md
• review_guide.md
• materials/ source folders
• materials_manifest.md
• source_brief.md
• workflow.md

Step 2: Sort material by signal strength

Rank sources before extracting:

1. exams, quizzes, assignments
2. lecture transcripts or lecture notes
3. slides and syllabus
4. group chats, Q&A logs, office-hour notes
5. professor bio, homepage, publication summaries

Use higher-signal sources to determine exam and review content. Use lower-signal sources to sharpen persona and identity.

When source files have been placed into materials/, always run the single-command build pipeline:

python ${CLAUDE_SKILL_DIR}/tools/build_professor_outputs.py "<professor-dir>"

This pipeline must:

• extract parseable text from pdf, pptx, docx, and text files into exports/extracted/
• refresh materials_manifest.md
• refresh source_brief.md
• generate persona.md, course.md, and review_guide.md
• validate the workspace before claiming it is ready

Read materials_manifest.md, source_brief.md, and the highest-signal extracted files first.

If ${CLAUDE_SKILL_DIR} is unavailable in the runtime, resolve tool paths relative to the skill root directory rather than the caller's working directory.

Step 3: Build three artifacts

Always generate these three files or sections:

• persona.md
• course.md
• review_guide.md

If the user explicitly wants it, also generate:

• mock exam
• likely key points
• oral-style explanation notes
• teacher-style chat replies

When updating existing artifacts:

• preserve strong evidence already reflected in the files
• replace [fill me] placeholders with concrete content
• keep unsupported claims marked as low-confidence inference

Step 3.5: Refuse fake confidence

If validate_professor.py warns that there are no exams, no transcripts, or no indexed sources, you should still help, but explicitly lower confidence and explain which parts are inferred.

Step 4: Keep the humor disciplined

Humor should come from recognition, not random jokes.

Prefer these patterns:

• "这题上课讲过" energy
• vague but familiar teacher phrasing
• passive-aggressive reminders
• overlong slides, underspecified key points
• exam warnings that feel suspiciously real

Avoid:

• insulting the professor
• fabricated misconduct
• fake official notices
• humor that reduces usefulness

Legal And Content Guardrails

• Treat imported materials as potentially sensitive by default.
• Do not encourage users to upload or redistribute content they do not have the right to use.
• Do not present generated text as an official notice, grading rule, or statement from the real professor.
• Do not fabricate private facts, misconduct claims, or internal school policies.
• If the user appears to be using private chats, recordings, unpublished materials, or other potentially restricted content without permission, warn briefly and continue only with clearly lawful, minimal assistance.
• When uncertainty exists, prefer summarization, study guidance, and low-confidence caveats over imitation that could be mistaken for the real person.

Output Requirements

persona.md