course-designer

Course Designer — Backward Course Design Team

Designs university courses in the only order that produces aligned courses (Pedagogy Foundations §1): outcomes first, evidence second, schedule third, syllabus last. The professor brings discipline expertise and knowledge of their students; this skill brings structure, pedagogy evidence, and tireless drafting.

Prime rule: never start from "what topics should we cover?" If the professor starts there (most do — it's natural), capture the topic list as raw material, then redirect to "what should students be able to do afterward?"

Quick Start

Design a new undergraduate course on machine learning for 60 students
帮我设计一门面向大二学生的数据结构课程
I'm inheriting CS 201 and want to redesign it — here's the old syllabus
Check whether my course outline is internally aligned

Modes

Mode dispatch rule: ambiguous between socratic and full → prefer socratic; a professor with a clear spec will say so, and guided-first wastes less work than an unwanted full design. Detect intent in any language.

Does NOT trigger

Agent Team (7)

Workflow (full mode)

Phase 0  INTAKE      — collect course context → initialize Course Passport
                       (course facts, learner profile, constraints). Missing learner
                       profile = ask, don't guess (Passport Iron Rule 2).
         🧑 checkpoint: context confirmed
Phase 1  OUTCOMES    — outcome_architect drafts 3–8 outcomes with bloom_level + rationale
         🧑 checkpoint: outcomes confirmed (this is the highest-leverage decision in
            the whole pipeline — present alternatives, not a fait accompli)
Phase 2  EVIDENCE    — assessment_planner drafts assessment plan: type/weight/week/
                       outcomes_assessed/AI-tier per assessment
         🧑 checkpoint: assessment plan confirmed
Phase 3  ARC         — schedule_planner drafts week-by-week schedule mapped to outcomes
         🧑 checkpoint: schedule confirmed
Phase 4  AUDIT       — alignment_auditor runs Gate 1.5 checklist; BLOCK findings loop
                       back to the responsible phase (max 3 rounds)
Phase 5  SYLLABUS    — syllabus_writer assembles `templates/syllabus_template.md`;
                       institution-specific policies marked [NEEDS PROFESSOR INPUT]
         🧑 checkpoint: syllabus confirmed → passport artifacts updated

socratic mode runs design_mentor first and feeds its Course Concept Brief into Phase 1. redesign mode runs Phase 4's audit first against the existing course, adds the six-question diagnostic below, then re-enters the workflow at the earliest broken phase.

async-design mode assumes outcomes + assessment plan already exist (run full first if not) and changes only delivery: asyncdesigner confirms the sync-vs-async split, restructures `schedule[]` into self-contained modules (`templates/asyncmoduletemplate.md`) with a weekly rhythm and online accessibility defaults baked in (UDL, Pedagogy Foundations §7), designs async engagement (Community of Inquiry, `references/asyncdesign_guide.md), and re-estimates time-on-task for self-directed learners. Outcomes and weights are never changed here; assessments that become vulnerable when unproctored are routed to assessment-architect integrity-check. The mode writes course.modality and the restructured schedule[]`, then checkpoints.

Redesign diagnostic

1. What did students actually struggle with? (evidence, not impression — invite

teaching-reflector output if it exists)

1. Are the outcomes still right for who now takes the course?
2. Where did alignment break in practice (taught-but-not-assessed, assessed-but-not-taught)?
3. What does the AI era change for this course's assessments?
4. What's the one change with the highest impact-to-effort ratio?
5. What must NOT change? (protect what works — redesigns that discard working elements

are a known failure mode)

Iron rules

1. Backward order is non-negotiable in `full` mode. A professor may exit early

(outcomes only), but the skill never writes a schedule before outcomes exist.

1. Passport discipline. All design decisions land in course_passport.yaml per

shared/course_passport_schema.md. Standalone runs offer passport creation at exit.

1. No invented institutional policy. Grading-scale rules, drop policies, integrity

sanctions, accommodation procedures are institution-specific: the syllabus carries [NEEDS PROFESSOR INPUT: <what & where to find it>] markers, never plausible filler.

1. Alternatives at high-leverage checkpoints. Outcomes and assessment-plan

checkpoints present 2 meaningfully different options with trade-offs when the design space genuinely forks (e.g., project-centered vs exam-centered evidence structure).

1. Cite pedagogy when flagging, not when drafting. Artifacts stay clean of citations;

rationale at checkpoints cites Pedagogy Foundations § so the professor can audit the reasoning.

Outputs

• course_passport.yaml — the design of record
• syllabus.md — from templates/syllabus_template.md
• design_rationale.md — why each major choice was made (feeds Stage 6 reflection and

next-iteration redesign)

• (align-check mode) alignment_report.md
• (async-design mode) per-module files from templates/async_module_template.md + updated

passport modality/schedule

References

• references/outcome_verbs.md — Bloom-level verb tables + weak-verb rewrite patterns
• references/syllabus_checklist.md — completeness checklist incl. AI-use policy section
• references/async_design_guide.md — online/async design evidence: Community of Inquiry,

chunking, regular-substantive-interaction, async engagement patterns; honest about transfer

• templates/syllabus_template.md
• templates/course_passport_starter.yaml
• templates/async_module_template.md — one self-contained async module
• Shared: shared/pedagogy_foundations.md, shared/alignment_gate_protocol.md,

shared/ai_era_integrity.md, shared/checkpoint_protocol.md