onboard-projectlisted
Install: claude install-skill Recluse/HyperMnesia
# Onboard a repository to HyperMnesia + Serena
Six steps. Five are mechanical; **step 3 is the one that decides whether any of this is
worth it**, so budget your attention there.
The end state: before the agent edits `src/api/routes.py` it already knows the rules for that
file (no search, no model), it can find the doc that explains the auth flow, it remembers what
you decided last month — and it asks a language server, not a vector index, where `foo` is
defined.
## 0. Check the prerequisites
```bash
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -c 'select 1' # store reachable
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -c 'select count(*) from components' # schema loaded
curl -s localhost:11434/api/tags | head -c 80 # embedder alive (Ollama; TEI: /health)
```
`-v ON_ERROR_STOP=1` on **every** psql call here, including the ones below: without it psql
prints the error to stderr and still exits 0, so a failed statement reads as an empty result —
the exact shape of failure this whole system exists to avoid.
If the schema is missing:
`psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -f sql/schema.sql -f sql/schema_mem.sql`.
If there is no store at all, `./hm init` brings up the Docker stack (Postgres + TEI, every
published port bound to `127.0.0.1`, a generated password in `deploy/docker/.env`), loads both
schemas, and prints the `DATABASE_URL` / `EMBED_BACKEND` lines to export.
Pick the repo tag now and use it everywhere — it is the scoping key for the whole tier a