memory-ingest-codelisted

# memory-ingest-code This is a **living document.** Mark steps as `[x]` and strike them through when done so subsequent runs skip them. See [`docs/skill-conventions.md`](../../docs/skill-conventions.md). The full module is at `memoryvault_kit/ingest/code_repo.py`. --- ## What this is — and what it's not > **The kit's code understanding ≠ Cursor's index.** > > Cursor indexes every line of source so it can write code. The kit is a > *memory*, not a code-gen tool. You don't need every function signature, > every implementation detail, every comment. You need to know: > > - **What products live in this repo** (structural understanding, captured once) > - **Who's working on what** (ongoing — captured from PRs) > - **What changed recently** (ongoing — captured from PRs) > - **What was the reasoning** (captured from PR descriptions, not the code itself) > > So this skill does two passes: > > 1. **Initial structural pass** — one-time, reads README + directory layout > + product config to map the surface. This is "the codebase's shape." > > 2. **Ongoing PR pass** — captures merged PRs as memories. Title + > description + files-changed-paths. **Never source contents.** This > is "what's happening in the codebase." > > Together those give you an *engineer's memory* — context for "who's > working on the SDK area," "what's the latest on the auth subsystem," "why did > we ship that caching change" — without ever flowing source code through > an LLM. > > If you want the kit to