Teach Your Agent to Teach with the using-tutor Skill

Where this fits

A developer asks their agent to "explain how database indexes work" and gets a crisp but shallow answer—yet what they actually wanted was to build a durable mental model. The using-tutor skill activates only when learning intent is explicit (like "开始学习英语" or "teach me Raft properly") and stays silent for ordinary Q&A, so the agent knows exactly when to switch from answering to teaching.

Why agents benefit

• Intent gate: Distinguishes explicit learning requests from routine factual questions, so the agent doesn't over-engineer simple answers.
• Silent control plane: No narration of status checks, tool calls, or skill switching—the learner only sees the teacher, keeping the interaction natural.
• Non-sycophantic correction: The Prometheus persona directly flags wrong reasoning with evidence or first principles, avoiding the flattery that plagues most LLM tutoring.
• Session recovery: Pending or interrupted Tutor work can be resumed, so long learning journeys survive context switches.
• Concise by design: Replies stay focused on learning outcomes, not Tutor mechanics or orchestration details.

Practical scenarios

Language learning with explicit intent

A user types "开始学习英语" in the chat. The agent recognizes this as explicit learning intent and starts a Tutor session immediately—no confirmation prompt, no Q&A mode. The user gets structured, teacher-led language practice without friction.

Deep technical concept mastery

A developer wants to truly understand how a distributed consensus algorithm works, not just get a summary. They say "teach me Raft properly." The agent enters Tutor mode, builds the mental model step by step, and corrects misconceptions as the developer reasons aloud.

Resuming an interrupted learning session

A user was mid-lesson on Python decorators, then switched tasks. When they return and say "continue where we left off," the skill recovers the pending Tutor work and picks up from the exact point of interruption.

Add it to your agent workflow

Install the skill:

npx skills add JanYork/llm-wiki-cli --skill using-tutor

Then reference it in your agent configuration for learning-intent detection:

skills:
  - name: using-tutor
    trigger: explicit learning intent or active Tutor session
    skip: ordinary factual questions

Compare before adopting

Before committing, evaluate on four axes:

• Quality signals: 47 GitHub stars is modest—check README depth, issue responses, and whether the author actively addresses feedback.
• Maintenance freshness: Look at the last commit date and open issues to confirm it isn't stale.
• Alternatives: Generic prompt templates for tutoring may cover simpler cases; using-tutor adds the silent control plane and session recovery.
• Workflow fit: If your agents only handle quick Q&A, this skill adds overhead. If you build education, language, or training agents, it's a strong fit.

Why it is worth tracking

The skill's design philosophy—a silent control plane with a non-sycophantic teaching persona—addresses a real gap in agent tutoring quality. At 47 stars it's early, but the focused scope and clear README signal an author who understands the problem. Track it if you build learning-oriented agents; evaluate it when your users ask for durable understanding rather than quick answers. For a curated view of similar agent skills, see OpenAgentSkill.