Teach Me Claude

Teach Me Claude

A Cowork / Claude Code plugin that teaches you to work with Claude by doing — a learning companion that delivers practical challenges, coaches you through them, grades the results against explicit rubrics, learns your workflows as you go, and tracks your journey.

• Free tier (shipped here): Series 1 — Foundations, 10 challenges taking a non-technical professional from "never worked with an AI agent" to confident, safe, effective delegation in Cowork. Complete on its own: the learner graduates with a working kit — their first skill (one that Socratically grills them until the brief is complete), brief template, verification checklist, delegation map, agent memory file, and a machine they commissioned themselves.
• Full curriculum: 100 challenges across 10 series — four knowledge-work domain packs (Research, Analytics, Documents, Presentations), then workflows, sharing, going public, automation, and shipping real software (roadmap/ROADMAP.md). Series 2–10 will be paid challenge packs (€20/series, €100 for all).

How it works

The plugin is the learner's companion:

• CLAUDE.md — the companion's operating contract: coach, never doer; no yes-machine — Socratic, honest, kind in that order; adapt to the learner's profession; grade honestly; track progress truthfully.
• Skills (invoked as /teach-me-claude:<name>, or triggered naturally in conversation):
  • teach-me — onboard a new learner (workspace setup, profile interview) or resume the journey
  • challenge — deliver a challenge: lesson taught conversationally (failure-first where designed), task set with parameters, time budget, and visible rubric
  • review — inspect the learner's actual artifacts against the rubric; pass/refine with specific feedback; debrief ritual (reflection, workflow profile, domain preview); record progress
  • progress — the journey map (text or HTML widget), the kit so far, and the full-curriculum roadmap
• Widgets (widgets/) — HTML templates the companion instantiates from real progress data: a journey map and a series graduation certificate.
• Resource guide (RESOURCES.md) — curated external pointers (Anthropic's skills repo, knowledge-work plugins, official docs), surfaced one at a time where a debrief calls for it.
• SessionStart hook — when a session opens inside a learning workspace, the companion greets the learner by name with where they left off.
• State — .teach-me/progress.json in the learner's own workspace folder (plain JSON, learner-owned, survives plugin updates; nothing stored elsewhere). Includes the learner profile, the accumulating workflow profile, per-challenge status, kit tracking, and reflections.

Each challenge file (challenges/series-01/) carries: Lesson, Task, Parameters, Rubric, and Companion notes — plus a Time budget (5–10 min openers, 15–20 core, longer capstones) and, where relevant, a domain Vehicle previewing the paid packs. The companion teaches from these — it never pastes them.

Install (for trial users)

From a clone of this repo:

claude --plugin-dir /path/to/teach-me-claude

or add the plugin through your Cowork plugin settings pointing at this directory / its future marketplace entry. Then say "I want to learn Claude" or run /teach-me-claude:teach-me.

Requirements: a Claude paid plan (Cowork or Claude Code). No technical background needed — that's the point.

Trial protocol (v0.2)

What we want to learn from trial users:

1. Does onboarding land — do they reach a first win inside one session?
2. Do the time budgets hold (especially the 5–10 minute opener)?
3. Does Challenge 4 deliver — does the grill-me skill get built, fire, and get used in later challenges?
4. Where do learners stall (challenge number, refine loops, drop-off)?
5. Does rubric grading feel fair and motivating?
6. Does the kit + roadmap framing generate pack interest (ask at exit)?

Ask trial users to share their .teach-me/progress.json and impressions after a week. Known v0.2 caveats — notably: verify plugin CLAUDE.md auto-load and workspace-skill installation mechanics on the tester's Cowork build; the skills are self-sufficient regardless.

Repository layout