rubysmithing

Full SIFT Protocol QA audit with structured rubric evaluation. Runs SIFT assessment (senior-qa-engineer) → rubric generation (director-of-ai-risk) → compliance evaluation (compliance-guardrail-agent).

Check or warm the rubysmithing gem API cache. Resolves current method signatures via Context7 MCP and persists results to the SQLite cache.

End-to-end Ruby diagnosis and fix. Runs Gemba Walk / root cause analysis (ruby-diagnostics-engineer) then applies targeted convention-aligned fixes (maintenance-architect).

Generate comprehensive documentation — YARD API docs (developer-experience-engineer) then wiki pages, guides, and README sections (agentic-software-engineer).

Implement a feature end-to-end. Verifies gem APIs (context-engineer) → implements (agentic-software-engineer or agentic-data-engineer based on feature type) → quality gate (senior-qa-engineer).

Use when designing or implementing data infrastructure for Ruby AI systems — clause-level SFL schema design, pgvector migrations, neuro-symbolic chunking pipelines, hybrid retrieval, knowledge graph architecture, or Ohm hot-storage models. Runs context-engineer as prerequisite.

Use when any Ruby development request needs routing to the rubysmithing skill suite — code generation, scaffolding, AI/NLP components, TUI interfaces, refactoring, QA assessment, or YARD documentation. Entry point that delegates to the appropriate specialized sub-agent.

Use when generating convention-aware Ruby code outside scaffold, genai, TUI, refactor, report, or yardoc domains — classes, modules, Rake tasks, config wiring, boot layers, data pipelines, POROs, error hierarchies, parallel workers, content parsers, or Gemfile decisions.

Use when building AI/NLP components in Ruby — LLM chatbots, tool-calling agents, RAG pipelines, vector search, embeddings, DSPy reasoning modules, MCP servers, local inference, NLP processors, or structured output. Runs context-engineer as prerequisite.

Use this agent to evaluate Ruby artifacts against a YAML evaluation specification produced by director-of-ai-risk. Applies rubric dimensions and checklist items with file:line evidence citations. Called internally by senior-qa-engineer (SIFT QA verification) and maintenance-architect (do-and-judge retry loop) — not user-invocable directly. Produces structured verdicts with PASS/FAIL and actionable retry feedback.

Ruby-targeted analysis skill implementing Gemba Walk (code archaeology, docs-vs-reality gaps), Muda waste analysis (dead methods, N+1 queries, unused gems, over-engineering), Root-Cause Tracing (backward call-chain from symptom to source), and Five Whys (iterative causal drilling). Use when user asks "why is this failing", "trace this bug", "root cause", "what's wasting cycles", "dead code audit", "gemba", "muda", "pre-refactor investigation", or reports a Zeitwerk error, circuit_breaker trip, or slow query. Auto-selects method from context. Produces findings keyed to refactor pattern names for direct handoff.

Gem API verification sub-skill for Ruby development. Automatically activates on first mention of any Ruby gem — especially ruby_llm, sequel, async, bubbletea, dspy.rb, pgvector, huh, dry-schema, circuit_breaker, fast-mcp, or informers — and resolves current method signatures and usage examples via Context7 MCP before code is generated. Results are cached for the session and persisted to SQLite across sessions. If Context7 resolution fails or is rate-limited, falls back to the tiered degradation protocol (stale cache → gem-registry ID → WARNING block) rather than silently guessing. Pairs with plan, genai, and tui as a prerequisite step.

Schema-first data infrastructure skill for Ruby AI systems. Activates on any mention of: PostgreSQL schema design, pgvector, HNSW index, chunking pipeline, embedding strategy, neuro-symbolic RAG, SFL schema, knowledge graph, clauses table, tokens table, clause_complexes, vault_chunks, Ohm model, Redis cache, symbolic metadata, ideational metafunction, transitivity, process type, participant role, register metadata, field tenor mode, entity extraction for storage, sequel migration, content hashing, idempotent pipeline, SHA256 deduplication, WordNet graph, formal concept lattice, hybrid search, RRF, Reciprocal Rank Fusion, hybrid retrieval, clause-level embedding, morphology tagging, POS tagging, dependency parsing, ruby-spacy tokens, lemma, pos_tag, tag_, dep_label, is_stop, is_punct, linguistic forms, IdeationalSchema, SFL annotator, ruby_llm-schema, informers reranker, taxis, clause complex, document ingestion, PDF extraction, DOCX parsing, OCR, kreuzberg, extract_file_sync, HierarchyConfig, batch extraction, document ingestor. Distinguishes schema design (DDL + migration), pipeline scaffolding (executable Ruby), and advisory (rationale + annotated snippet) modes. Always runs rubysmithing-context as a prerequisite step before generating any library-specific code. Single focused file output; specs only on explicit request.

AI/NLP component scaffolder and advisor for Ruby projects. Activates on any mention of: LLM chat, chatbot, agent, tool-calling, function calling, RubyLLM::Agent, RubyLLM::Tool, streaming, RAG, retrieval augmented generation, vector search, semantic search, embedding, pgvector, DSPy, reasoning module, chain of thought, ReAct, MCP server, MCP tool, model context protocol, ruby_llm, RubyLLM, NLP, entity extraction, named entity recognition, sentence segmentation, chunking, document processing, local inference, informers, spaCy, ruby-spacy, prompt pipeline, or structured output. Detects whether request is scaffolding (generate file) or advisory (explain approach). Always runs rubysmithing-context as a prerequisite step before generating any library-specific code. Single focused file output; specs only on explicit request.

Convention-aware Ruby code generator for the terminal-native AI orchestration stack (dotenv, tty-config, zeitwerk, async, dry-schema, sequel, journald-logger, circuit_breaker, parallel, concurrent-ruby). Use for generating Ruby classes, modules, Rake tasks, config wiring, boot layer code, data pipelines, POROs, error class hierarchies, parallel processing workers, content parsers, and Gemfile decisions. Applies project-detected conventions (RuboCop, StandardRB, or community idioms) automatically. For single-file output under ~50 lines or pure stdlib work, activates Lite Mode — no dry-schema, no async, no circuit breakers. Multi-file scaffold requests always use Standard Mode regardless of per-file line count. Does NOT handle: LLM/AI/NLP code (genai), TUI interfaces (tui), refactoring (refactor), quality reports (sift), or YARD documentation (yardoc). Always defers to context-engineer for gem API verification before generating library-specific code.