learnlisted

# /learn — persist reusable insight Turn one-off discoveries into **durable knowledge** so the next session — or another developer — does not re-derive them. This is a **Guide** half of the harness trust loop (it informs future work). Pairs with [`retro`](../retro/SKILL.md) (which surfaces what is worth learning) and [`harness-propose`](../harness-propose/SKILL.md). > **Authority**: [`AGENTS.md`](../../../AGENTS.md) is binding. This skill **MUST NOT** modify any file without > explicit user approval. Knowledge augments context; it does not override rules. > **Reference loading (progressive disclosure):** > Read [principles](reference/principles.md) first, then use the [KB index](reference/kb/INDEX.md). > Load only topic files matching the current task; MUST NOT recursively load the entire KB. ## Preprocessing (MUST) - Read `CLAUDE.md` to understand the current rule/skill structure. - Read `.claude/knowledge/README.md` to learn the existing knowledge layout (and avoid duplicates). ## Procedure ### 1. Mine the conversation Find **non-obvious** insights from the current work: - a newly discovered coding/test pattern or rule; - a bug's root cause and its fix; - a project-specific gotcha (e.g. an intentional decision that looks like a bug); - the *why* behind an architecture/decision. Skip the obvious — anything already in `AGENTS.md`, a skill, git history, or `.claude/knowledge/`. ### 2. Check existing docs Confirm whether a home already exists: `AGENTS.md`, the releva