# CRM suite — the skill-of-skills composition manifest. # It COMPOSES the seven base crm/* skills into ONE DataK3 bucket: shared master data, # 36 disjoint tables, and a SINGLE deferred graph assembly. It is not a fork and it does # not re-spec any workflow — the mechanics live in the seven base contracts; this manifest # is the ordered install DAG + the master-data wiring + the one-graph rule + the union Q&A. # The human walkthrough is posts/enterprise_software/crm/suite.md (published id crm-suite). # Lives at skill/enterprise_software/crm/suite.skill.yaml; served over MCP at dodil://skills/suites/crm; # executed by `dodil skill scaffold suites/crm`. check:skills binds it to its post. id: suites/crm version: 1.0.0 summary: >- Scaffold a complete, customizable CRM on ONE DataK3 bucket — master data (accounts/contacts/leads/ opportunities/activities) + six workflows (lead-to-opportunity, qualification-scoring, pipeline-forecast, quote-cpq, campaign-to-lead, account-360) + one relationship graph, assembled once. 36 disjoint tables, shared masters joined across skills, no ETL and no second copy. module: crm workflows: [master-data, lead-to-opportunity, qualification-scoring, pipeline-forecast, quote-cpq, campaign-to-lead, account-360] pillars: [sql, graph, vector, ignite, models] industry: software # cross-industry base; overlays (crm/overlays/*) reshape per industry overlays: [saas, manufacturing, finserv, real-estate] # Binding to the narrative (check:skills verifies these resolve + agree) post: posts/enterprise_software/crm/suite.md route: /library/crm-suite # It composes the seven base skills (no overlay by default). An overlay is an additive diff on top. composes: skills: [crm/core, crm/lead-to-opportunity, crm/qualification-scoring, crm/pipeline-forecast, crm/quote-cpq, crm/campaign-to-lead, crm/account-360] overlays_apply_on_top: [crm/overlays/saas, crm/overlays/manufacturing, crm/overlays/finserv, crm/overlays/real-estate] # BUILDING A SUBSET (not all seven). You do NOT have to install the whole suite to get one workflow. subsets: rule: >- To build a SUBSET, install crm/core FIRST — it owns the five masters (accounts/contacts/leads/opportunities/activities) + the vectors + crm_node/crm_edge that every workflow reads — then install only the workflow skills you want. Each workflow skill's `consumes.master` lists its master dependencies (e.g. qualification-scoring needs leads/contacts/activities/opportunities; pipeline-forecast needs opportunities/accounts); crm/core provides them all. standalone_without_core: >- Each workflow skill ALSO ships a `when: standalone` stub_masters step that creates just the masters IT reads — so a single workflow runs on an empty bucket with no crm/core. But once you want TWO workflows sharing the same masters, install crm/core once instead of letting each stub its own (avoids divergent seed data). graph: >- Skip the graph entirely (no CREATE GRAPH) unless you install crm/account-360 — the single deferred CREATE GRAPH belongs to it (the last contributor). crm/core's standalone family-rollup graph is only built when you run crm/core alone with hierarchy=true. # One question set for the whole suite. Only the shared/composition-answered ones collapse here; # per-workflow policy knobs stay (they are genuine best-practice decisions, defaulted by the overlay). params: bucket: { type: string, default: crm, prompt: "Bucket name? (ONE bucket for all seven skills)" } seed_data: { type: enum, options: [demo, empty], default: demo } # Acme family + Greyparrot demo, shared by every skill's Test overlay: { type: enum, options: [none, saas, manufacturing, finserv, real-estate], default: none } product_pitch: { type: string, default: "an AI data platform for enterprise teams" } # shared: lead-to-opportunity + qualification-scoring gates hierarchy: { type: boolean, default: true } # core: crm_node/crm_edge + subsidiary_of edges (feeds the single graph) framework: { type: enum, options: [bant, meddic], default: bant } # qualification-scoring attribution_model: { type: enum, options: [first_touch, last_touch, linear, multi_touch], default: linear } # campaign-to-lead # ORDERED install DAG — masters first, the graph's last contributor last. This is the scaffold order. order: - crm/core # 1. owns the 5 masters + 3 vectors + crm_node/crm_edge; the system of record - crm/quote-cpq # 2. owns products (account-360 whitespace needs the catalog) - crm/lead-to-opportunity # 3. discovery warehouse + sequence engine; converts orgs -> core leads/opportunities - crm/qualification-scoring # 4. scores core.leads (writes lead_scores, reads core masters) - crm/pipeline-forecast # 5. stages/forecast over core.opportunities - crm/campaign-to-lead # 6. campaigns/attribution over core.opportunities; may add campaign graph edges - crm/account-360 # 7. LAST graph contributor -> runs the single deferred CREATE GRAPH # Shared master data — owner skill -> the sibling skills that JOIN it (no copy; one row, many readers). # Derived from each base contract's provides.owns_master -> siblings' consumes.master. master_wiring: accounts: { owner: crm/core, consumers: [crm/lead-to-opportunity, crm/pipeline-forecast, crm/quote-cpq, crm/campaign-to-lead, crm/account-360] } contacts: { owner: crm/core, consumers: [crm/lead-to-opportunity, crm/qualification-scoring, crm/quote-cpq, crm/campaign-to-lead, crm/account-360] } leads: { owner: crm/core, consumers: [crm/lead-to-opportunity, crm/qualification-scoring, crm/campaign-to-lead] } opportunities: { owner: crm/core, consumers: [crm/lead-to-opportunity, crm/qualification-scoring, crm/pipeline-forecast, crm/quote-cpq, crm/campaign-to-lead, crm/account-360] } activities: { owner: crm/core, consumers: [crm/lead-to-opportunity, crm/qualification-scoring, crm/account-360] } organizations: { owner: crm/lead-to-opportunity, consumers: [] } products: { owner: crm/quote-cpq, consumers: [crm/account-360] } campaigns: { owner: crm/campaign-to-lead, consumers: [] } # THE single CREATE GRAPH — run exactly once, after the last contributor. This is the composition subtlety. graph_assembly: graph: crm_graph rule: >- In suite mode every graph-touching skill ONLY inserts into crm_node/crm_edge — NO skill runs CREATE GRAPH. The snapshot rule (crm-on-dodil.md:170 / crm-core.md:215) means edges added after CREATE are invisible, so the suite defers the single CREATE GRAPH crm_graph NODES (crm_node KEY id) EDGES (crm_edge SRC src DST dst) to account-360, the last contributor. Node id ranges are pre-partitioned (accounts 1-9999, contacts 10000-99999, campaign nodes above) so contributors never collide on id. contributors: - { skill: crm/core, edges: [subsidiary_of, works_at] } - { skill: crm/campaign-to-lead, edges: [member_of, influenced], when: { multi_touch_graph: true } } - { skill: crm/account-360, edges: [partner_of, competes_with, supplies, owns_product], runs_create_graph: true } # LIVE-FOUND composition subtlety (2026-09-02): once account-360's partner_of edges join the single # graph, the standalone core family-rollup graph_khop('crm_graph',1,2,'in') OVER-COUNTS (it walks ALL # incoming edges regardless of rel — Greyparrot's partner_of edge leaks into the Acme family, $66k not # $36k). The suite family rollup MUST constrain the traversal to rel='subsidiary_of'. Type your edges. # The ordered scaffold plan an agent executes (tools resolve against dodil://commands via check:skills). steps: - id: bucket title: One bucket for the whole CRM when: always tools: [data_bucket_create] detail: Create the single DataK3 bucket {{bucket}}. Every one of the seven skills lands here — one data plane. - id: core title: Install crm/core (the system of record) when: always tools: [data_table_create, data_table_upsert, data_pg] detail: >- 5 masters (accounts/contacts/leads/opportunities/activities) + 3 VECTOR(2048) tables + crm_node/crm_edge. In suite mode core populates node/edge but DEFERS CREATE GRAPH. Seed the Acme family demo if seed_data=demo. - id: quote-cpq title: Install crm/quote-cpq (products before whitespace) when: always tools: [data_table_create, data_table_upsert] detail: products/price_books/price_book_entries/quotes/quote_lines/discount_policy — the catalog account-360 reads. - id: lead-to-opportunity title: Install crm/lead-to-opportunity (discovery -> convert into core) when: always tools: [data_table_create, data_table_upsert, ignite_app_deploy] detail: >- organizations/org_embeddings + the flow/email tables + conversions. Converts write straight into core's leads/opportunities (shared masters) — no re-declaration; product_pitch is the shared gate input. - id: qualification-scoring title: Install crm/qualification-scoring (scores core.leads) when: always tools: [data_table_create, data_table_upsert, ignite_app_deploy] detail: lead_scores/scoring_policy; reads core.leads + core.opportunity_vectors. framework/threshold are the tunable knobs. - id: pipeline-forecast title: Install crm/pipeline-forecast (over core.opportunities) when: always tools: [data_table_create, data_table_upsert, data_sql] detail: pipeline_stages/forecast_snapshots/deal_risk; weighted Σ amount×probability over the shared opportunities master. - id: campaign-to-lead title: Install crm/campaign-to-lead (attribution over core.opportunities) when: always tools: [data_table_create, data_table_upsert, data_pg] detail: >- campaigns/campaign_members/touchpoints/attribution; attribution JOINs core.opportunities. If multi_touch_graph, projects campaign nodes/edges into crm_node/crm_edge (still deferred — no CREATE here). - id: account-360 title: Install crm/account-360 + run the SINGLE graph assembly (LAST) when: always tools: [data_table_create, data_table_upsert, data_pg, data_bolt] detail: >- relationships/account_summary/whitespace. Adds partner_of/competes_with/supplies/owns_product edges, THEN the one CREATE GRAPH crm_graph over the fully-populated crm_node/crm_edge. Family rollup filters rel='subsidiary_of'. - id: overlay title: Apply an industry overlay (optional) when: { overlay: [saas, manufacturing, finserv, real-estate] } tools: [data_pg, data_table_create, data_table_upsert] detail: >- Apply crm/overlays/{{overlay}} — an ADDITIVE diff on top of the 36 tables (ALTER-add columns via the pg wire + maybe one extra table + gate/edge tweaks). Never rewrites a base skill. See the four overlay contracts. # The union Q&A (the composition removes redundant asks — proposal §10.1). customize: - param: bucket ask: "One bucket name for the whole CRM?" effects: any: "all seven skills scaffold into this single bucket — one data plane, cross-skill JOINs are free." - param: seed_data ask: "Load the shared Acme demo dataset, or ship empty schemas?" effects: demo: "one demo (Acme family + Greyparrot + the planted near-dup) drives EVERY skill's Test — asked once, not seven times." empty: "empty schemas across all 36 tables; each skill's Test switches to structural assertions." - param: overlay ask: "Which industry? (drives all four overlay diffs at once — subscriptions/dealers/households/listings + gate defaults)" effects: none: "the cross-industry base suite." saas: "subscriptions + ARR/seat columns + renewal forecast lane + product-qualified gates." manufacturing: "dealers + install_base + territory-weighted graph + quote-to-order." finserv: "households + referrals + KYC hard gate on the MEDDIC scorer." real-estate: "buyer↔listing vector matching + commission splits." - param: product_pitch ask: "What are you selling, in one line? (shared by lead-to-opportunity + qualification-scoring — asked ONCE)" effects: any: "the same pitch feeds both the discovery qualify gate and the lead scorer — composition removes the duplicate ask." - param: [framework, attribution_model, hierarchy] ask: "The per-workflow policy knobs — qualify framework (BANT/MEDDIC), attribution model, family rollups?" effects: any: "genuine best-practice decisions the overlay may default but you tune; not removed by composition." # Composition-answered asks REMOVED (documented so the union is auditable): # - lead-to-opportunity's "hand-feed leads or discover?" consumes leads/opportunities FROM crm/core — no re-ask. # - account-360's "seed a product catalog for whitespace?" — products come from crm/quote-cpq (step 2). # - qualification-scoring's opportunity_vectors — come from crm/core, not re-created. # - campaign-to-lead's "project campaign nodes into crm_graph?" — crm_node/crm_edge are owned by crm/core; the # suite's single deferred CREATE GRAPH covers it (no per-skill graph create in suite mode). # What proves the ASSEMBLY (the suite's Test is about composition, not re-testing each skill). tests: - "data table list -> 36 tables coexist, NO name collision (34 SQL/vector + crm_node + crm_edge)" - "cross-skill JOIN: lead_scores (qualification-scoring) JOIN leads (crm/core) -> Jane qualified 82, Sam disqualified 34" - "cross-skill JOIN: attribution (campaign-to-lead) JOIN opportunities (crm/core) -> Q3 Webinar attributed $24,000 to opp-labs, ROI 4x" - "single graph: one CREATE GRAPH crm_graph after account-360; partner_of edge (node 4->1) traversable via Bolt" - "family rollup filtered rel='subsidiary_of' -> $36,000 (Labs 24k + EU 12k); naive graph_khop('in') over-counts to $66k once partner_of joins" tested_branches: - { seed_data: demo, overlay: none, tested_at: 2026-09-02 } # default suite — full single-bucket assembly, live-validated