oc-importing-claude-memorylisted

# Importing-Claude-Memory — one-shot migration of CC native memory into OC A repo that previously used Claude Code accumulated a native per-project memory at `~/.claude/projects/<slug>/memory/` (a directory of one-fact-per-file notes + a `MEMORY.md` index). The OpenCode harness can't see it (hermetic rule), and it holds operator/decisional knowledge **not derivable from code or the project `CLAUDE.md`**. This skill migrates the durable, non-sensitive subset into the OC project-root `MEMORY.md`, **once**, by hand. ## Identity & boundary (non-negotiable) - **Operator-invoked only.** Never auto-run. `oc-surveying-codebase`, `build`, and the delivery loop NEVER invoke this. The hermetic rule ("never read `~/.claude`") stays literally true for everything else — this is the one named exception, and it only fires when the human explicitly runs it. - **Read-only on `~/.claude`, scoped to `projects/<slug>/memory/` only.** Never read any other `~/.claude` path. **Never write to `~/.claude`** (no "mark as imported" on the CC side — Finding 7). Enforce "once" with an OC-side marker only (below). ## Model — run the classification on a STRONG model This skill runs **once per repo** and its hardest step (judging sensitivity + staleness of foreign notes) is exactly where a cheap open-source model fails open. So do NOT run the classification on the default hands. Use a **strong model — `openai/gpt-5.5`** (or the strongest available). The one-shot cost is negligible against the stakes (a