Llm wiki crm

LLM Wiki CRM

CRM layer on top of the PM wiki. Tracks relationship health, account status, feature asks, and auto-enrichment for people and company entities. Complements Salesforce — this layer captures PM-specific context that CRMs don't: product feedback, feature asks, communication history, strategic relationship signals.

When This Skill Activates

• "relationship health", "who haven't I talked to", "dormant accounts", "check relationships"
• "enrich [person/company]", "auto-enrich [person/company]", "fill in [person/company] details" (This skill enriches relationship/CRM context — tier, touchpoint, feature asks. For factual/web enrichment of a stub page, use llm-wiki-research's Auto-Research.)
• "account health", "strategic accounts", "how are my accounts"
• "what do customers want", "feature asks", "what are accounts asking for", "customer asks"
• "log touchpoint with [person/company]", "update [entity] relationship", "set [entity] to strategic"
• "CRM", "key accounts", "contact cadence", "overdue contacts"

New CRM Frontmatter Fields

These fields extend the core wiki SCHEMA.md. Add them to entity pages as relevant. See skills/llm-wiki-crm/templates/SCHEMA-crm-fields.md for a merge-ready patch.

Person Entity Additions

relationship_tier: strategic | active | watch | dormant
last_touchpoint: YYYY-MM-DD
meeting_cadence: daily | weekly | biweekly | monthly | quarterly | ad-hoc
next_meeting: YYYY-MM-DD        # optional
influence_level: high | medium | low
enriched_at: YYYY-MM-DD         # when auto-enrichment last ran

Company Entity Additions

relationship_tier: strategic | active | watch | dormant
account_health: green | yellow | red
last_touchpoint: YYYY-MM-DD
key_asks: []                    # list of feature/product asks from this account
arr_tier: enterprise | mid-market | smb | prospect   # optional, no dollar figures
enriched_at: YYYY-MM-DD         # when auto-enrichment last ran

Tier Definitions

Tier	Meaning
strategic	High-impact relationship. Flag if silent > 14 days.
active	Regular engagement. Flag if silent > 30 days.
watch	Relationship at risk or low-signal. Flag if silent > 60 days.
dormant	No active engagement. Informational only.

Account Health Definitions

Value	Meaning
green	Healthy engagement, no known issues
yellow	Some friction — risk of churn, unresolved asks, reduced engagement
red	Needs immediate attention — escalation risk, silent, or major open issue

Wiki Path Resolution

WIKI=$(cat .wiki-path 2>/dev/null | tr -d '[:space:]')
WIKI=${WIKI:-${CLAUDE_PLUGIN_OPTION_wiki_path:-${WIKI_PATH:-$(pwd)}}}

Operations

1. Relationship Health Check

Trigger: "relationship health", "who haven't I talked to", "dormant accounts", "check relationships"

① Scan all person and company entity pages for last_touchpoint: and relationship_tier:. ② Apply staleness thresholds (strategic >14d, active >30d, watch >60d, dormant = skip). ③ Surface grouped by tier:

"3 strategic contacts need attention: [[X]], [[Y]], [[Z]] (last contact: N days ago)" ④ For each flagged entity: last touchpoint, key_asks, account_health, next_meeting if set. ⑤ Offer to run Pre-Meeting Briefing (core skill §9) for any flagged entity. ⑥ Log: ## [date] crm-health | flagged: N strategic, N active

---

2. Auto-Enrichment

Trigger: "enrich [entity]", "auto-enrich [person/company]", "fill in [entity] details"

Scope — avoid overlap with llm-wiki-research: this operation is a lightweight factual pass whose real job is populating relationship/CRM fields (relationship_tier, last_touchpoint, account_health, key_asks). For a deep factual/web sweep of a company or person, defer to llm-wiki-research's Auto-Research when it's installed — run that first, then layer the CRM fields on top. The steps below are the standalone fallback when research isn't available.

Company enrichment: ① WebSearch: company name + "about", funding, headcount, key products, recent news ② Extract: founded year, HQ, employee count, funding stage, key products, recent press ③ Update entity page. Set enriched_at: today. Add inline: [source: <url>, <date>] ④ Bump coverage: stub → partial if meaningful data found ⑤ Flag time-sensitive data with confidence_decay_days: 90

Person enrichment: ① WebSearch: person name + company + role, LinkedIn (public), recent talks/posts ② Extract: current role, company, background, public positions on relevant topics ③ Update entity page. Only public information — never infer private details. ④ If person is strategic tier and no persona page exists, offer to create one.

---

3. Account Health Dashboard

Trigger: "account health", "strategic accounts", "how are my accounts"

① Scan company entity pages with relationship_tier: strategic | active. ② Aggregate: account_health distribution, key_asks frequency across accounts, last touchpoint age. ③ Output:

• 🔴 Red accounts: N
• 🟡 Yellow accounts: N
• 🟢 Green accounts: N
• Top feature asks: [feature] (N accounts), [feature] (N accounts) ④ Surface cross-account key_asks patterns — high-signal roadmap inputs. ⑤ Offer to file as queries/account-health-<date>.md.

---

4. Feature Ask Tracker

Trigger: "what do customers want", "feature asks", "what are accounts asking for"

① Grep all entity pages for key_asks: frontmatter. ② Aggregate: count asks per feature/theme across accounts. ③ Cross-reference with roadmap-tagged pages — is the ask already on roadmap? ④ Output table:

Feature ask	Accounts requesting	On roadmap?
[feature]	N accounts	yes / no / unknown

⑤ File as queries/feature-asks-<date>.md.

---

5. Update Relationship Fields

Trigger: "log touchpoint with [person/company]", "update [entity] relationship"

① Read entity page. ② Update: last_touchpoint: to today, relationship_tier: and account_health: if specified. ③ Prompt for notes to add to interaction history section. ④ If meeting just happened, offer to ingest transcript (hands off to core §2).

---

Behavioral Notes

• Verify from source, don't launder wiki prose (core skill → Session Defaults → Provenance-tier & falsification guard): account health, targeting, pipeline, and "what does this account want" answers are decision-bearing. The authoritative source is the primary system (CRM/SFDC, the warehouse, Gong), not wiki prose — route there first. Tier every asserted claim computed/primary/recalled, never generalize from a couple of named accounts onto the whole book without testing it against the full dataset, and run a falsification pass before any exec-facing account synthesis. The CRM layer's own fields (relationship_tier, account_health, key_asks) are recalled state — confirm them against live touchpoints before acting on them.
• No dollar figures: arr_tier uses tiers only. Never record ARR, deal sizes, or revenue figures.
• Private by default: every wiki page is private unless it carries shareable: true (core skill privacy model). Do NOT add private: true — it is noise. Account intel pages simply stay unmarked and are excluded from exports automatically.
• Enrich only public data: never infer private details (financials, internal decisions, personal info).
• key_asks is a list: normalize phrasing across accounts so aggregation works cleanly.
• Log every CRM operation with crm- prefix:
  • ## [date] crm-health | flagged: N strategic, N active
  • ## [date] crm-enrich | <entity> | enriched_at: <date>
  • ## [date] crm-touchpoint | <entity> | last_touchpoint: <date>
  • ## [date] crm-feature-asks | filed: queries/feature-asks-<date>.md