Consolidate 35 skills into 10 umbrella skills

[pyramation]

Summary

Reorganizes 35 standalone skill directories into 10 umbrella skills to reduce cognitive load for AI agents navigating this repo. Each umbrella skill has a SKILL.md that acts as a router with a reference table, and absorbed skills become files in the umbrella's references/ directory. Also deletes planning-blueprinting (not ours) and updates AGENTS.md / CLAUDE.md with the new structure.

The 10 umbrellas:

Umbrella	Absorbs
constructive	safegres, services-schemas, deployment, server-config, functions, cnc-execution-engine, constructive-env
pgpm	github-workflows-pgpm (as new reference)
constructive-graphql	graphql-codegen, sdk-search, graphql-query
constructive-starter-kits	boilerplate-pgpm-init, boilerplate-nextjs-app, boilerplate-authoring
constructive-frontend	constructive-ui, crud-stack, meta-forms
constructive-testing	pgsql-test, drizzle-orm-test, supabase-test, drizzle-orm, pgsql-parser-testing
constructive-ai	pgvector-rag, github-workflows-ollama
constructive-tooling	constructive-pnpm, inquirerer-cli, readme-formatting
graphile-search	(stays as-is)
fbp	fbp-types, fbp-spec, fbp-evaluator, fbp-graph-editor

Absorbed files were moved/renamed only — no content modifications. The new umbrella SKILL.md files are the only net-new authored content.

---

Updates since last revision: constructive-ai SDK-first refactor

Rewrote the constructive-ai skill to use Constructive SDK + codegen'd ORM patterns only, removing all generic RAG/pgvector tutorials that taught patterns outside the platform (e.g., new Pool(), raw pgpm docker start setup).

Changes:

• Rewrote SKILL.md — now shows the "Constructive AI Flow" (SDK provision → codegen → ORM query → RAG), with inline code examples using db.field.create, db.index.create, and db.document.findMany with vectorEmbedding filter
• Deleted 4 generic reference files: pgvector-rag.md, rag-embeddings.md, rag-similarity-search.md, rag-setup.md — these contained standalone Pool-based TypeScript and generic pgvector tutorials
• Renamed 2 files: rag-agentic-kit.md → agentic-kit.md, rag-ollama.md → ollama.md
• Created pgvector-sql.md — consolidated raw SQL reference with every code block prefixed -- SQL to clearly distinguish from SDK patterns
• Rewrote rag-pipeline.md — end-to-end RAG using SDK provisioning, codegen, ORM ingestion/retrieval, and Ollama generation (includes complete RAGService class)

Final reference file set: rag-pipeline.md, ollama.md, ollama-ci.md, pgvector-sql.md, agentic-kit.md

---

Updates since last revision: agentic-kit rewrite + RAG reference for constructive-graphql

agentic-kit.md fully rewritten to match the actual agentic-kit source. The old version was significantly outdated — it used new Pool(), referenced a fictional RAGProvider class, imported from @sf-ai/agentic-kit, and pointed to deleted skills.

New agentic-kit.md covers:

• Correct API: AgentKit manager, OllamaAdapter, AnthropicAdapter, OpenAIAdapter
• Factory helpers: createOllamaKit(), createAnthropicKit(), createOpenAIKit(), createMultiProviderKit()
• Correct packages: agentic-kit, @agentic-kit/ollama, @agentic-kit/anthropic, @agentic-kit/openai
• Streaming, embeddings via OllamaClient, GenerateInput options, AgentProvider interface
• 3-pass RAG architecture (query routing → vector search via ORM → LLM synthesis) based on real patterns from agentic-db

Created constructive-graphql/references/search-rag.md — RAG-specific query patterns using the codegen'd ORM:

• Single-table and multi-table vector search
• Embedding ingestion via ORM .update() (not raw SQL)
• Hybrid search combining fullTextSearch + vectorEmbedding in a single where
• 3-pass RAG with LLM query routing

Updated cross-references:

• constructive-graphql/SKILL.md — added search-rag.md to reference table + cross-references to constructive-ai for agentic-kit and rag-pipeline
• constructive-ai/SKILL.md — added search-rag.md cross-reference, updated agentic-kit description

---

Review & Testing Checklist for Human

• Verify agentic-kit.md API matches actual agentic-kit source: The rewrite uses AgentKit, OllamaAdapter, AnthropicAdapter, OpenAIAdapter, and factory helpers (createOllamaKit, etc.) based on reading the source at packages/agentic-kit/src/index.ts. Confirm these exports, class names, and method signatures (generate, generateStreaming, addProvider, setProvider) are correct in the published package.
• Verify search-rag.md ORM patterns: Uses vectorEmbedding: { vector, metric, distance } in where, embeddingVectorDistance as a select field, and model.update({ where: { id }, data: { embedding } }) for ingestion. These patterns are based on agentic-db's search.ts and embed-all.ts — confirm they match the actual codegen'd ORM output.
• Check for RAG content overlap between skills: Both agentic-kit.md (in constructive-ai) and search-rag.md (in constructive-graphql) contain multi-table vector search and 3-pass RAG examples. Verify this duplication is acceptable or whether one should defer to the other more clearly.
• Verify constructive-ai/SKILL.md SDK code examples match actual API surface: The Quick Start shows db.field.create(), db.index.create(), and db.document.findMany({ where: { vectorEmbedding: { vector, metric, distance } } }). Confirm the ORM actually generates vectorEmbedding filter, EMBEDDING_VECTOR_DISTANCE_ASC orderBy, and embeddingVectorDistance select field with these exact names.
• Check cross-references resolve: constructive-ai links to ../constructive-graphql/references/search-rag.md, search-pgvector.md, and search-composite.md. constructive-graphql links to ../constructive-ai/references/agentic-kit.md and rag-pipeline.md. Confirm all linked files exist on this branch.

Suggested test plan: Load the constructive-ai skill and ask an agent to "build a RAG pipeline with Anthropic" — verify it routes to agentic-kit.md and produces AgentKit + AnthropicAdapter code (not the old RAGProvider / new Pool() patterns). Then load constructive-graphql and ask for "vector search for RAG" — verify it routes to search-rag.md with ORM-based examples.

Notes

• The old agentic-kit.md (513 lines) was a complete rewrite (96% changed) — it contained a fictional RAGProvider class, @sf-ai/agentic-kit imports, new Pool() database access, and references to 5 deleted skills. None of this matched the actual agentic-kit source.
• The search-rag.md RAG patterns (VECTOR_CONDITION helper, multi-table parallel search, embedding ingestion via .update()) are modeled on the real implementation in the agentic-db repo (packages/rag/src/search.ts, embed-all.ts).
• The old constructive-testing had framework-details.md and anti-patterns.md reference files that were deleted because the rewritten SKILL.md no longer references them. The anti-patterns content was folded into the new SKILL.md itself.
• constructive-graphql/references/codegen-options.md, codegen.md, search.md, pagination.md were created in a prior PR (Consolidate into unified constructive-graphql skill, delete old codegen + sdk-search #58) and are expected to already exist on the branch.
• Merged main into the feature branch to resolve conflicts; also removed a stale constructive-graphql-codegen.zip that came in from main (that skill no longer exists).
• The constructive-sdk-agentic-flow repo's instruction.md was verified — it already references the new umbrella skill names (constructive-graphql not constructive-graphql-codegen). No changes needed there.
• No CI exists on this repo, so validation is manual only.

Link to Devin session: https://app.devin.ai/sessions/9f711b5939654d0caec800c90bbd073a
Requested by: @pyramation

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring