rai-agent-skills

Take a built RelationalAI model to production — deploy it into a Snowflake schema and version it through the op log (branch, collaborate, merge, and tear down with the `rai models` CLI), or deploy it as a Snowflake CoWork (Cortex) agent. The one path-to-prod skill across deployment targets. Use when deploying a model, managing its deployed lifecycle, or operationalizing it as a Cortex agent — not for first-time install/connect (see rai-setup), building the model (see rai-ontology, rai-pyrel), or interpreting reasoner output.

Graph algorithm selection and execution on PyRel v1 models — construction from ontology patterns, parameter tuning, and result extraction. Use for questions about a network's structure — centrality and importance, community detection, connectivity and components, reachability and dependencies, shortest paths and distance, node similarity, and variable-length path enumeration (where the route itself is the answer).

PyRel v1 language — modeling syntax (concepts, properties, relationships, data loading), business rules as derived properties (validation, classification, tiers, flags), and query construction against `relationalai.semantics.Model` (selects, filters, joins, aggregates, export). Load BEFORE writing any PyRel code, even your first line — prior knowledge of the syntax is likely stale. Use whenever the user asks to model, load, derive, classify, flag, query, count, rank, aggregate, join, or export data from a RAI model, even if they don't say PyRel. Not for ontology design decisions (see rai-ontology), optimization formulation (see rai-prescriptive-problem), graph algorithms (see rai-graph-analysis), or GNN work (see rai-predictive-modeling).

Translation, ideation, and routing layer between an ontology and the RAI reasoners. Surfaces questions the data can answer, classifies them by reasoner family (prescriptive, graph, predictive, rules), and translates user-facing problem framings into the technical implementation hints the downstream reasoner skills need. Use before choosing a reasoner workflow or when scoping what to build next.

Guides diagnosis of RAI engine performance, failed transactions, CDC/data-stream health, and CDC engine management. Use when a reasoner is slow or queuing, a transaction or batch has failed, a CDC stream is suspended or quarantined, or CDC engine sizing/recovery is needed.

Builds and evolves RAI ontologies — greenfield starter builds from Snowflake tables or local data, and all domain-modeling decisions (concepts, relationships, identity, subtypes, data mapping, layering, enrichment). Use when creating a new RAI model, starting a proof of concept, onboarding a dataset, or reviewing and enriching an existing ontology. Authoring the PyRel itself (syntax, data loading, rules, queries) is `rai-pyrel`.

Build graph neural network (GNN) models — concepts, Snowflake data loading, task relationships, graph edges, and PropertyTransformer features. Use for node classification, regression, and link prediction tasks; for training, predictions, and evaluation, see `rai-predictive-training`.

Configure and train graph neural network (GNN) models, generate predictions, evaluate results, and manage trained models. Use when ready to train, generate predictions, evaluate, or manage models; for concepts, data loading, edges, and feature configuration, see `rai-predictive-modeling`.

Formulates optimization and constraint-satisfaction problems from ontology models — decision variables, constraints, objectives, problem-type classification, solver selection, global constraints, and pre-solve validation. Use when building, reviewing, debugging, or relaxing a formulation, through the point where a validated Problem and chosen solver are ready to run. Not for executing the solve or interpreting output — status, extraction, sensitivity, conflicts (see rai-prescriptive-results).

Runs optimization solves and interprets the output — solve execution and parameters, diagnostics requests (sensitivity, conflict / IIS), status codes, solution extraction, quality assessment, sensitivity analysis, infeasibility diagnosis, and stakeholder explanation. Use when executing a formulated problem or analyzing anything a solve produced — status, duals, marginals, conflicts, trivial solutions, what-if scenarios. Formulation changes and solver selection go back to rai-prescriptive-problem.

Setup and configuration for RelationalAI — first-time install walkthrough and all raiconfig.yaml tuning. Use when installing RAI, connecting to Snowflake, running RAI locally on DuckDB for development, or editing raiconfig.yaml. Not for writing PyRel model code (see rai-pyrel) or solver usage and diagnostics (see rai-prescriptive-results).

Reviews RAI agent skills for structure, content quality, prompt engineering, boundaries, examples, and agent usability. Use when creating, reviewing, or auditing skills in rai-agent-skills or rai-agent-skills-private.

Bumps the plugin version across manifest files, commits, and creates a local git tag for rai-agent-skills. Pushing and publishing the GitHub release happen separately, under human review. Use when cutting a release.