Skip to main content

/learn-grade

Grade an explanation against a named concept and its normalized corpus. The skill asks for concrete gaps, misconceptions and transfer evidence; an LLM score alone does not establish mastery.

/learn-grade --concept-id borrow-checker --learner-id learner-1 --corpus-path /path/to/corpus.json --explanation "My explanation..." --goal-id rust-basics

The corpus supplies key points and known misconceptions. The resulting grade includes overall_score, misconceptions_absent, gap_list and transfer_score. The explanation gate requires overall_score >= 0.7 and misconceptions_absent == 1.0; transfer and delayed retention remain separate gates.

The skill instructs the agent to screen its draft with sycophancy-correction S-02 when available, or use the documented manual fallback. Its write-grade.sh helper persists grade JSON; it does not enforce an independent grader or automatically rewrite incorrect feedback. Record the actual grader and screening evidence. See Anti-Sycophancy.