v1.2.0 — Correctness, guardrails, coverage & adoption

v1.2.0 — Correctness, guardrails, coverage & adoption (2026-06-15)

A large release implementing a verified 8-dimension improvement audit (42 findings, each
premise-checked). Closes the gap between the suite's promises and its mechanisms.

Correctness (bugs)

• Fixed a real pipeline deadlock: workload_audit had no writer, so Gate 1.5 check D1
  (BLOCK) would deadlock on every real run. schedule_planner now persists it; the gate
  stays read-only.
• Added schema fields that were used-but-undefined: term_calendar, learner_profile.cohort_evidence.
• iteration_history single-writer discipline; removed an orphaned directory.

Honesty guardrails (prose → structure)

• The recommendation letter (the suite's self-described "worst single failure") now has a
  template with a [C#] claim-to-intake trace appendix — a fabrication shows as a blank
  source cell. Feedback comments likewise template-backed.
• check_passport.py P11: structural PII guard (no student email/ID/roster in the passport).
• check_content_markers.py: [NEEDS PROFESSOR INPUT]/[VERIFY] enforced by Gate T4.
• The four-block checkpoint presentation is now a required emission format; a per-artifact
  output self-check (shared/output_self_check.md).

Machine-checkable layer

• Gate 3.5 is now executable (check_quality_gate.py) for its decidable core.
• Registry lint extended (R6–R9) to plugin/marketplace versions, description counts, and
  the ARCHITECTURE version — drift is now CI-enforced.
• Main-repo internal link checker (check_links.py).

Coverage (+5 modes) — gradebook/grade-distribution analytics, graded group projects +
peer assessment, accommodation operationalization, async-course design, and a carefully
bounded suspected-misconduct (integrity-case) companion.

Adoption — render_document.py (Markdown → DOCX/PDF) and export_lms.py
(GIFT / QTI 2.1 / Common Cartridge), so artifacts reach the LMS and the printer.

Pedagogy & credibility — honest caveats on the Bloom/active-learning/cognitive-load
claims; new sections on alternative grading and signature (studio/clinical/performance)
pedagogies; a second worked showcase (HIST 247 humanities seminar) with inspectable
person-affecting outputs (recommendation letter + intake trace + feedback report).

Suite: 15 skills · 89 modes · 76 agents. 63 tests; CI enforces gates, links, manifests.

---

Install: /plugin marketplace add YujxZJCN/teaching-skills → /plugin install teaching-skills. Codex: teaching-skills-codex.

v1.1.2 — Enriched course dashboard

A visual upgrade to the course dashboard (/ts-dashboard). Still one self-contained HTML file, no dependencies — now much richer.

New in the dashboard

• Course at a glance — three visual cards: a grade-composition donut (pure-CSS conic-gradient, flags unallocated weight), Bloom-distribution bars across all six levels, and a semester strip (one cell per week: blue when materials are built, striped for logistics weeks, red dots for assessments due, topic on hover)
• Alignment matrix — the constructive-alignment triangle made visible: an outcomes × assessments grid with ✓ cells colored by Bloom level, plus a taught-weeks strip under each outcome
• Policies section — grading / late / AI-use / integrity / attendance rendered from the passport, with [NEEDS PROFESSOR INPUT] markers highlighted so unfilled institutional facts can't be missed
• Richer detail — assessments show authoritative due-weeks from the schedule; the artifact ledger shows confirmed/missing-on-disk counts; the learner profile surfaces prior knowledge, known difficulties, and cohort-analyst aggregates when present; the workload audit shows a gauge plus its arithmetic constants
• Usability — every LO/A/W chip is a link that jumps to and auto-expands its section; sections are collapsible; a print stylesheet is included

No skill modes or passport fields changed. Regenerate anytime: python3 scripts/build_dashboard.py course_passport.yaml. 28 tests green in CI.

v1.1.1 — Learner analysis (学情分析) + course dashboard

Two additions: evidence about your students flows in, and a complete view of your course flows out.

New skill: cohort-analyst (学情分析)

Turns professor-held student data — ability lists, pre-course diagnostics, pre-lesson questionnaire results — into evidence-based teaching decisions. 5 modes:

• instrument (/ts-diagnostic) — designs the ungraded diagnostic itself: prerequisite probes, two-tier misconception items, labeled self-report items; every item must inform a named teaching decision before deployment
• cohort-profile (/ts-cohort) — analyzes results into per-concept readiness distributions, misconception prevalence, and heterogeneity, with mandatory instrument-strength and small-N caveats
• lesson-calibration — reteach/activate/skip recommendations, misconception-targeted distractors, pacing fixes — each traceable to a specific finding
• grouping — goal-matched, pseudonymized grouping plans with rotation rules (and a refusal of learning-styles pseudoscience; differentiation runs on prior knowledge, which has evidence)
• progress — cohort-level trajectory re-analysis mid-term

Privacy architecture: the unit of analysis is the cohort. Only aggregates enter the Course Passport (shown verbatim before writing); individual-level questions route to student-mentor under its person-affecting rules; no predictive risk scores, no tracking labels.

New: course dashboard

python3 scripts/build_dashboard.py course_passport.yaml renders the passport into a single self-contained HTML file — open it in any browser:

• gate status, stored and live-recomputed (a stale green is impossible)
• outcomes with Bloom badges and coverage; assessment plan with weight bars and integrity status
• week-by-week schedule with clickable links to the local artifact files (missing files flagged visibly)
• artifacts ledger with confirmation status, workload audit, iteration history

New teaching-pipeline mode dashboard (/ts-dashboard); regenerate at any checkpoint — the passport is the source, the dashboard is a build product. See the showcase dashboard built from the example course.

Suite: 15 skills · 84 modes · 71 agents · 29 commands. Registry lint + 27 tests green in CI.

v1.1.0 — Machine-checkable contracts

The Course Passport's contracts are now executable, not just documented — following advice from the ARS author (discussion #417) that the deterministic layer is what compounds.

New

• shared/course_passport.schema.json — formal JSON Schema for the Course Passport (required fields, enums, id patterns)
• scripts/check_passport.py — P1–P10 cross-reference invariants the schema can't express: id uniqueness, the assessed_by ↔ outcomes_assessed and taught_in ↔ schedule.outcomes bidirectional mirrors, referential integrity, weights summing to 100
• scripts/check_alignment_gate.py — Gate 1.5 (constructive alignment, checks A1–D3) executed deterministically instead of interpreted by a model; honors previously dismissed findings; PASS / PASS-WITH-WARNINGS / FAIL verdicts
• scripts/check_registry_consistency.py — the repo's three-places rule (SKILL.md ↔ MODE_REGISTRY.md ↔ filesystem) made executable
• Test suite + CI — the showcase passport serves as a golden fixture that must always pass clean; 24 mutation tests assert each check fires on targeted breakage; GitHub Actions runs everything on every push

Changed

• gate_runner_agent, passport_keeper_agent, and alignment_auditor_agent now run the validators when Python 3 is available and degrade gracefully to manual checklist evaluation (stated as lower assurance) when it isn't
• READMEs (EN + 中文) and ARCHITECTURE document the new layer

No skill modes or passport fields changed — existing passports validate as-is.

v1.0.0 — Teaching Skills for Claude Code

Initial public release: a comprehensive Claude Code skill suite for university professors covering the full teaching lifecycle — design → build → assess → deliver → reflect → improve.

What's in v1.0.0

Core pipeline (7 skills)

• course-designer — backward course design: Bloom-tagged outcomes → assessment plan → semester arc → syllabus; Socratic mode for blank-page starts
• lesson-builder — lesson plans, lecture notes, slide outlines, a 15-technique active-learning catalog, cases, discussion guides
• assessment-architect — blueprint-first exams, rubrics, TILT project briefs, AI-resilience integrity audits, item analysis
• student-mentor — feedback, intervention plans, recommendation letters, difficult communications (evidence-bound, draft-only)
• submission-auditor — compile your template/requirements into a checkable spec, audit submissions with located evidence, batch fairness pass
• teaching-reflector — bias-honest evaluation analysis, mid-course feedback, portfolio/statement, SoTL design
• teaching-pipeline — orchestrates everything over the Course Passport with two non-skippable gates and a weekly delivery loop

Extensions (7 skills)

• deck-studio (renders real slide decks: Marp/Pandoc/Beamer/python-pptx), lab-forge (executable labs, per-student datasets, autograders), media-scripter (6–9 min video scripts, captions), course-publisher (announcements, weekly emails, course site, LMS packaging), ta-coordinator (handbooks, grading calibration, allocation), accreditation-mapper (outcome–standard matrices with evidence status), bilingual-courseware (glossary-bound EN⇄中文 materials)

Foundations

• Course Passport — single source of truth; any fresh session resumes from it
• Alignment Gate (1.5) + Quality & Integrity Gate (3.5) — deterministic checks, professor acknowledgment
• 12-principle evidence-based pedagogy reference; AI-era integrity model (policy tiers + structural resilience, no detection tools)
• Honesty by design: [NEEDS PROFESSOR INPUT] / [VERIFY] markers, person-affecting outputs evidence-bound and never auto-sent

14 skills · 78 modes · 67 agents · 27 /ts-* commands · English + 简体中文 (README.zh-CN.md)

Install

/plugin marketplace add YujxZJCN/teaching-skills
/plugin install teaching-skills

Architecture inspired by Academic Research Skills (Cheng-I Wu).