relationship-intelligence

name	relationship-intelligence
description	Query and maintain Sunil's relationship graph. Use when the task involves who someone is, family context, outreach, reconnects, anniversaries, intros, Rhaine handoff, or relationship follow-through.
allowed-tools	Bash, Read, Write

Relationship Intelligence

This skill is the native OpenClaw path for Linus to reason about Sunil's network.

Use it when the user asks things like:

• who is a person
• who Sunil should reach out to
• whether Linus knows Sunil's parents, wife, EA, or family context
• intros, reconnects, anniversaries, birthdays, or follow-ups
• travel or logistics where people context matters
• what arrived recently on WhatsApp, who sent it, or what recent WhatsApp context matters
• what recent Slack or email context matters
• what important emails are waiting on Sunil
• what a Drive doc, Roam note, or meeting history says about a person or company

Do not rely on generic memory retrieval first. Query the relationship store directly.

Canonical files

• Store: /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite
• Script: /root/.openclaw/workspace/relationship-intel/relationship_intel.py
• Radar markdown: /root/.openclaw/workspace/memory/RELATIONSHIP-RADAR.md
• Operating state: /root/.openclaw/workspace/memory/HOT-STATE.md
• Promoted docs: /root/.openclaw/workspace/memory/PROMOTED-DOCS.md
• Index: /root/.openclaw/workspace/memory/PEOPLE-INDEX.md
• Person pages: /root/.openclaw/workspace/memory/people/*.md

Fast paths

Weekly or proactive relationship guidance

Use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  brief --json

This is the default path for:

• "who should I reach out to this week?"
• "what open loops do I have with people?"
• "what birthdays or anniversaries matter?"

If Linus needs the smallest high-value surface first, use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  operating-state --json

Person lookup

Use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  summary "Prasad Rao" --json

Use this before asking obvious questions about:

• parents
• Angelic / Emile / family
• Rhaine
• contacts, intros, prior relationship context

Broad search

Use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  search "founder toronto" --json

Recent WhatsApp context

Use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  messages \
  --channel whatsapp \
  --days 2 \
  --direction inbound \
  --limit 20 \
  --json

If the store is empty or stale, refresh it first:

/root/.openclaw/workspace/relationship-intel/whatsapp_history_sync.sh

For broad “what important messages came in recently?” questions, use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  channel-brief \
  --channel whatsapp \
  --days 2 \
  --limit 12 \
  --json

Other imported sources

Use messages for channels that were imported as message evidence:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  messages \
  --channel slack \
  --days 14 \
  --limit 20 \
  --json

Use docs-search for Roam, Drive, and Calendar evidence:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  docs-search "Tribble reasoning" \
  --channel roam \
  --json

Use gmail-guided when the task is email-shaped but should be driven by people/open-loop/project context rather than a broad mailbox crawl:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  gmail-guided \
  --person "Rohan Verma" \
  --objective "investor introductions" \
  --days 365 \
  --json

Use import-himalaya-email when the question depends on the personal Gmail account that is still accessed through Himalaya:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  import-himalaya-email \
  --account-name personal \
  --folder "Needs Action" \
  --folder "EA to Action" \
  --max-messages 200 \
  --json

For proactive inbox triage or “what am I neglecting?” questions, use:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  email-attention \
  --days 21 \
  --limit 12 \
  --json

If the question is specifically about deals, intros, partnerships, customers, or investor follow-through, bias the ranking:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  email-attention \
  --days 21 \
  --focus deals \
  --limit 12 \
  --json

Incremental freshness

For routine cross-inbox freshness on the host, use:

/root/.openclaw/workspace/relationship-intel/incremental_sync.sh

To inspect what it will do before running:

DRY_RUN=1 /root/.openclaw/workspace/relationship-intel/incremental_sync.sh

Record a real touchpoint

When Sunil mentions a meaningful interaction, update the graph:

python3 /root/.openclaw/workspace/relationship-intel/relationship_intel.py \
  --db /root/.openclaw/workspace/relationship-intel/relationship_intel.sqlite \
  touch "Dev Rajendran" \
  --note "Caught up on Slack about next week and the UiPath thread." \
  --channel slack \
  --memory-dir /root/.openclaw/workspace/memory

Then refresh the prompt surface:

/root/.openclaw/workspace/relationship-intel/relationship_sync.sh

Operating rules

• Query the relationship store before asking the user obvious people-context questions.
• For recent WhatsApp questions, use messages first instead of guessing or saying you cannot inspect WhatsApp.
• Never use the generic message tool to read WhatsApp/Slack/email history. That path is for delivery actions, not archival review. Use channel-brief or messages from relationship_intel.py.
• For Slack or email recency questions, use messages --channel slack|email before improvising.
• For targeted inbox questions about a known person, project, or open loop, use gmail-guided before broad Gmail import in the live Google lane.
• For important-inbox triage, use email-attention before improvising from generic email search.
• For deal or intro slippage, use email-attention --focus deals so internal ops, family, and accepted calendar noise do not dominate the ranking.
• Prefer incremental_sync.sh for routine Gmail/Calendar/Drive/personal-email freshness instead of broad manual re-imports.
• Personal Gmail ([e-mail]) currently comes from Himalaya; corporate Gmail ([e-mail]) comes from Google OAuth. Treat them as one evidence plane once imported.
• For company or personal note questions, use docs-search before claiming the context is unavailable.
• Prefer operating-state and HOT-STATE.md before pulling the broader document archive into prompt context.
• After any major multi-source import, run reconcile-identities so duplicate people do not linger under separate phone/email records.
• Treat brief and summary as the primary source; treat markdown pages as a quick human-readable fallback.
• If the relationship store is thin or ambiguous, say so and ask only the missing high-signal question.
• Handle fast person lookups and reconnect queries inline; do not spawn workers just to answer a trivial relationship question.
• Use workers only when the task turns into broader research, parallel outreach preparation, or heavier cross-source synthesis.
• Do not auto-send outreach. Use the relationship context to draft or prepare a handoff.
• When travel, logistics, or booking work involves known people, pull their summaries first so Linus can infer context intelligently.