/memory

description: Search OpenLoomi memory (files + knowledge base + insights) — thin doorway into the openloomi-memory sub-skill
argument-hint: "<query>"
allowed-tools: Bash(node ${CLAUDE_PLUGIN_ROOT}/skills/openloomi-memory/scripts/openloomi-memory.cjs *)

/openloomi:memory <query>

Thin doorway into the [`openloomi-memory`](../skills/openloomi-memory/SKILL.md) sub-skill.

• With a `<query>` argument → runs `search-all` across local memory

files, the knowledge base (RAG), and recent insights.

• With no argument → lists the most recent 7 days of insights

(`list-insights --days=7`) so the user can see what's already been captured.

For everything else (filter by channel, get a single document / insight, create / update / delete, time-travel queries, living connections, entity registry) say "memory search", "knowledge base", "insights" — the sub-skill's frontmatter triggers on those phrases and walks Claude through the right subcommand.

The CLI lives at `${CLAUDE_PLUGIN_ROOT}/skills/openloomi-memory/scripts/openloomi-memory.cjs` and auto-reads the bearer token from `~/.openloomi/token`.

Steps

1. **With query** — run a comprehensive search:

   node ${CLAUDE_PLUGIN_ROOT}/skills/openloomi-memory/scripts/openloomi-memory.cjs \
     search-all "<query>"

Aggregate results from `memory` (local files), `knowledge` (RAG), and `insights`. Print top 5 from each source with file path / doc id / insight id and a short preview. If the user wants more, run the same with `--limit` raised, or fall through to a specific source via the sub-skill's `search-memory` / `search-knowledge` / `list-insights` commands.

2. **Without query** — list recent insights:

   node ${CLAUDE_PLUGIN_ROOT}/skills/openloomi-memory/scripts/openloomi-memory.cjs \
     list-insights --days=7

Print the most recent 10 (or all if fewer than 10) with their `type`, `groups`, `time`, and a short `content` preview.

3. If the user asks for a specific source after seeing results, dispatch to the matching subcommand:

| Source | Subcommand | | -------------- | -------------------------------------------------------------------------------- | | Local files | `search-memory "<query>" [--directory=<subdir>]` | | Knowledge base | `search-knowledge "<query>"` or `list-documents` / `get-document <id>` | | Insights | `list-insights [--days=N] [--channel=<gmail\|telegram\|...>] [--keyword=<k>...]` |

Failure modes

| Symptom | What to surface | | ------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------- | | `~/.openloomi/token` missing or unreadable | Run `/openloomi:setup` first to mint a guest bearer. | | CLI exits with `AUTH_FAILED` | Same as above — token is stale or invalid. | | `search-knowledge` returns empty | Likely no documents uploaded yet. Tell the user the knowledge base is empty and they can upload via OpenLoomi Desktop → Knowledge Base. | | `search-memory` returns empty | Local `~/.openloomi/data/memory/` has no matches for the query. That's a legitimate result, not an error — surface the empty `results` array. |

Constraints

• **Never** delete memory files or insights without explicit user

confirmation of the target path / id. The CLI's `delete-memory` / `delete-insight` commands are irreversible.

• **Never** write to `~/.openloomi/data/memory/` directly from this

command. Use `add-memory` (with `--file` and `--directory`) so the filename and placement are deterministic and audit-able.

• The skill is a **doorway** — don't duplicate the sub-skill's full

API reference here. If the user asks for something this doc doesn't cover, route to the sub-skill via natural-language triggers.