| name | customer-brief |
| description | One-command grounded account snapshot of a customer — KB profile, contacts, apps, recent meetings, document inventory, pending e-signatures, and recent email activity, synthesized into a compact brief. Use when {{OPERATOR_FIRST_NAME}} asks "where are we with X", "brief me on X", "catch me up on X", or "what's the latest with X" for a customer/client/company. |
| version | 1.0.0 |
| author | hermes-mini-setup (reference build) |
| license | MIT |
| platforms | ["linux","macos","windows"] |
| metadata | {"hermes":{"tags":["Customer","Brief","CRM","Knowledge Base","KB","Account"]}} |
Customer Brief
A grounded, read-only snapshot of one customer — the pre-meeting / pre-call "where are we with them" answer. Trigger on "where are we with X", "brief me on X", "catch me up on X", "customer brief for X".
Read-only skill: it never writes to the KB, never sends anything. No approval gates needed; just gather, synthesize, present.
Procedure
1. Deterministic gather
python ~/.hermes/skills/productivity/customer-brief/scripts/brief_data.py "<customer name/slug/alias>"
Returns one JSON object: customer (identity + metadata), people (contacts w/ roles, primaries first), apps, meetings (latest 10), kb_docs (recent 15 + total), pending_esign (docs out for signature matched to this customer's people). On error it lists the known customer slugs — if the ask plausibly matches one, retry with that slug; otherwise tell {{OPERATOR_FIRST_NAME}} who IS known and stop.
2. Enrich (both, in parallel where possible)
- KB content:
mcp_rag search scoped to the customer (customer=<slug>) with 2-3 queries shaped by what step 1 revealed — e.g. the latest meeting's topic, an active project/app, "proposal". You're after substance for the brief's "current threads", not a re-listing of titles.
- Email: search {{OPERATOR_FIRST_NAME}}'s mailbox (google-workspace gmail search) for recent threads with the customer's people — query their email addresses/domain, last ~30 days. Note who wrote last and whether anything looks unanswered.
3. Synthesize the brief
One compact Slack message, mobile-first (no wide tables). Shape:
- Header — company, stage/role relationship in one line (from KB profile metadata), since-date.
- People — name · role, primaries first, one line each.
- Latest activity — the 2-3 most recent meetings with a one-line takeaway each (pull the takeaway from the meeting doc content, not just its title), plus anything active found in email (who's waiting on whom).
- Open items — pending e-signatures (subject + sent-when), unanswered email threads, anything the KB content marks as blocked/next.
- Assets — apps (with repos) and the KB doc count, one line.
- Gaps — one line, honest: what the KB doesn't have (e.g. no meetings filed, profile thin), so {{OPERATOR_FIRST_NAME}} knows the confidence level.
Cite what came from where only when it matters (e.g. "per the 7/16 planning meeting"). Never pad; if a section is empty, drop it.
4. Offer, don't push
End with: offer a deliverable-export version (branded DOCX/PDF) if {{OPERATOR_FIRST_NAME}} wants it as a document — only produce it if he says yes.
Notes
- Resolution accepts name, slug, or alias; a fuzzy fallback catches partials ("acme" →
acme-construction).
pending_esign only includes signature requests matched to this customer's people by email/name — an empty list means none for THIS customer, not none at all.- Meetings and docs come from the structured tables (deterministic); narrative quality comes from your step-2 searches. Do both — the script alone under-informs the brief.
