contact-360

contact-360

"I'm about to call them in 10 min — who are they to me?" Aggregates every touchpoint with a person into one brief.

Verbs (the installed lark-* skills): contact +search-user (resolve name), contact +get-user (details by --user-id), im +messages-search, calendar +agenda, vc +search, docs +search, task +search, base +record-search. Mail has no search verb in raw CLI — use atomic lark-mail skill. Atomic command shapes + token flags: connectors/LARK-RECIPES.md.

Workflow

1. Resolve identity. contact +search-user --query "<name>" --jq '.data.users[]|{name:.localized_name,open_id,email}' → pick open_id (data is a dict {users,has_more}). Then contact +get-user --user-id <open_id> --jq '.data.user' for dept/title/manager.
2. Read memory. people.md (personal notes), priority-people.md (VIP tier), team-roster.md, accounts.md (sales).
3. Parallel fetch touchpoints — all output-projected:
   • IM: im +messages-search --sender <open_id> --start <t-90d> --end <t> --page-size 50 --jq '.data' (peek shape — needs search:message scope; CLI has no recipient-direction filter, sender-only; --start/--end need a timezone offset).
   • Mail: via atomic lark-mail skill (no raw mail search verb).
   • Meetings: vc +search --participant-ids <open_id> --start <t-90d> --end <t> --jq '.data.items[]|{id,info:.display_info}' (data is a dict — meetings in .data.items[]). Calendar +agenda has NO attendee filter — skip or scan agenda titles only.
   • Docs: docs +search --query "<name>" --jq '.data' (peek shape; needs search:docs:read scope; keyword match on name only).
   • Tasks: task +search --follower <open_id> --page-limit 5 --jq '.data.items[]|{summary,due:.due_at}' (data is a dict — tasks in .data.items[]; also try --assignee).
   • CRM (sales): base +record-search --base-token <t> --table-id <accounts> --json '{"keyword":"<name>","search_fields":["Owner"]}'.
4. Synthesize. Last interaction (most-recent IM/mail/meeting); recurring themes (top 3 keywords); open threads (meetings w/o follow-up, unanswered mail); past decisions (keyword match); outstanding commitments (tasks where person is follower); keyword count of "urgent/escalate/critical" (count only — no labels).

Output

👤 CONTACT 360 — <Name> · <Title> · <Dept> · Manager: <Manager>
📅 LAST INTERACTION — <date>: <type> · "<topic>"
🔄 RECURRING THEMES (90d) — <theme>: <N> · ...
⏳ OPEN THREADS — unanswered mail / meeting w/o action / stale task
✅ PAST DECISIONS (90d) — <date>: <summary> (source)
📌 COMMITMENTS — you owe / they owe
🧠 SUGGESTED OPENERS (3) — each anchored to a real thread
⚠️ NOTES FROM people.md — <if exists>
Action sau gặp: update people.md / tạo task follow-up / log decision

Hard rules

1. Read-only against Lark. Only memory/people.md may be appended (after explicit confirm) — not a Lark mutation.
2. Resolve names, never raw open_ids in output.
3. Truncate per source to top 5 — full brief <30 lines.
4. 0 IM + 0 mail + 0 calendar in 90d → warn "no recent touchpoints".
5. No psychoanalysis — count keyword occurrences, no sentiment labels.
6. CRM lookup only for sales role.
7. Suggested openers must reference actual data — never generic.

Edge cases

• Internal but new (<30d) → "first-time meeting — read team-roster.md".
• External (no open_id) → skip IM, use mail + calendar + CRM.
• Name collision (+search-user >1 hit) → prompt user to pick.
• User asks about self → redirect to morning-brief / weekly-review.

Memory

Required: do-not.md. Recommended: people.md, priority-people.md, team-roster.md, accounts.md (sales).

Integration

On-demand or auto from meeting-prep (1:1 / external attendee) and deal-update (pre-call). After meeting, offers to append a 1-2 line note to memory/people.md — closes the "remember context" loop.