/v-memory-refresh

description: (Re)index docs/superpowers prose into the local V-memory cache so recall is current. Incremental by file hash; runs fully offline (FTS5, pure stdlib). Optionally enable the semantic lane with a one-time bootstrap. Run it after pulling new docs, or when /v:remember looks stale.

You are running **`/v:memory-refresh`**. Args: `{{args}}`.

**Resolving the plugin root.** The `scripts/` this command calls ship with the plugin — they are not files in your own repository. Resolve the plugin root once per session before calling any of them:

CV="${CLAUDE_PLUGIN_ROOT:-$(ls -d "$HOME"/.claude/plugins/cache/*/superpowers-v/*/ 2>/dev/null | sort -V | tail -1)}"
CV="${CV:-$PWD}"; CV="${CV%/}"

`CLAUDE_PLUGIN_ROOT` is set for hooks but is not set in this Bash environment, so treat it as a hint, never the whole answer — the fallback line covers an installed plugin cache or a checkout of this repo.

**Default — offline, FTS5, no install, no network:**

python3 "$CV/scripts/compound-v-memory.py" refresh
python3 "$CV/scripts/compound-v-memory.py" doctor

Report the `doctor` summary (files / chunks / staleness / whether embeddings are bootstrapped).

**Semantic lane (opt-in).** Embeddings are OFF by default and live **outside the repo** (`~/.cache/compound-v/memory/<repo-id>/`). Enabling them is the **only** step that touches the network — and it must be explicit, never from a hook:

python3 "$CV/scripts/compound-v-memory.py" bootstrap                 # creates the out-of-repo venv + model (one time)
python3 "$CV/scripts/compound-v-memory.py" refresh --with-embeddings # populate vectors

If the project opted into embeddings at [`/v:init`](v-init.md) (`memory.embeddings: true` in `.claude/compound-v.json`), the engine **already** adds vectors on a plain `refresh` once bootstrapped — you don't need the flag. If `{{args}}` asks for `--with-embeddings` and `doctor` shows embeddings are not bootstrapped, run `bootstrap` first (tell the user it will download a ~200 MB model once). The semantic lane is **scale-gated**: it only changes ranking once the corpus is large enough to matter; on a small corpus FTS5 already wins. If bootstrap fails (offline / no wheels), the engine stays FTS5-only — recall still works. See [`skills/compound-v/memory.md`](../skills/compound-v/memory.md).