Learner (Grade A)安全测试数据 AI 能力：克劳德 AI 的 A 等品质。修复后知识提取 - captures 代码库特定的非明显知识，通过质量门槛

Do I need to manually classify each learning, or does Learner handle that?

Learner classifies each saved learning automatically based on its content: domain‑level expertise (why something works) or workflow details (how to perform a task). No manual tagging is required, though you may review the classification for accuracy.

Asked by Elif Yildiz · Aug 23, 2025

How does Learner store the extracted knowledge?

Accepted learnings are automatically categorized as either expertise or workflow knowledge and appended chronologically to the project's CLAUDE.md file, providing a searchable, version‑controlled record for future reference.

Asked by Chidi Okonkwo · Jul 31, 2025

What are the quality gates that determine if a learning is saved?

Learner applies three mandatory gates: (1) Not Googleable – the insight can’t be found with a quick web search; (2) Codebase Specific – it must rely on the project’s unique architecture or quirks; and (3) Required Real Effort – the discovery had to involve genuine debugging work. If any gate fails, the learning is not recorded.

Asked by Ingrid Bauer · Jul 10, 2025

When should I invoke the Learner skill during debugging?

Invoke Learner after resolving a non‑trivial bug or surprising behavior, especially if the debugging session involved more than three back‑and‑forth exchanges, the root cause was counter‑intuitive, the fix required understanding a project‑specific convention, or you uncovered an undocumented interaction.

Asked by Celia Ramirez · Jul 5, 2025