Claude Skills by Yakoub-ai

Complete pymc-marketing API reference for writing and reviewing MMM code. Use when you need exact constructor signatures, method names, parameter lists, or code patterns for pymc-marketing v0.19.1+. Load this skill when writing new model code, debugging import errors, checking method signatures, or reviewing code that uses pymc-marketing. Also activate when you need to look up plotting methods, evaluation functions, cross-validation setup, or transformation classes.

Channel attribution, ROAS calculation, contribution decomposition, and results interpretation for Marketing Mix Models. Use when extracting channel contributions, calculating return on ad spend, generating response/saturation curves, interpreting model outputs, assessing channel effectiveness, decomposing the target variable into components, or presenting MMM results to stakeholders. Also activate when the user asks about contribution shares, waterfall charts, or which channels are over/under...

Budget allocation optimization and sensitivity analysis for Marketing Mix Models. Use when optimizing media budget allocation, setting up BudgetOptimizer with CustomModelWrapper, defining budget bounds and constraints, running sensitivity analysis, comparing current vs. optimal allocation, or advising on budget reallocation strategies. Also activate when the user asks about optimal spend, budget constraints, or allocation scenarios.

Data quality assessment and preparation for Marketing Mix Models. Use when evaluating datasets for MMM readiness, checking data requirements, validating columns, assessing collinearity, handling missing values, or preparing marketing spend data for pymc-marketing modeling. Also activate when the user asks about minimum data requirements, data granularity, or feature engineering for MMM.

Convergence diagnostics, fit evaluation, and debugging for Marketing Mix Models. Use when checking rhat, ESS, divergences, evaluating model fit metrics (R-squared, MAPE, wMAPE), debugging sampling issues, assessing overfitting, running model comparison (LOO/WAIC), or validating cross-validation results. Also activate when the user encounters convergence warnings, poor model fit, or needs to understand why a model isn't working.

Catalog of external factors and control columns for MMM. Use when recommending controls for a new model, evaluating which external factors to include, or understanding why certain controls matter for a given industry or region.

Greenfield vs brownfield MMM decision guide. Use when the user asks about starting a new MMM from scratch vs improving an existing one, or when designing the modeling strategy for a company with prior MMM experience.

Intake questionnaire for MMM projects. Use when starting a new MMM project, resuming an incomplete intake, or updating the spec.yaml. Covers company context, target unit, channel inventory, controls, seasonality, and greenfield vs brownfield classification.

MMM iterative improvement mechanics: tournament-based model selection and posterior-informed prior tightening. Use when designing or running the improvement loop, understanding tournament scoring, debugging why improvement stalled, or explaining the refinement strategy to stakeholders.

Model construction and prior specification for Marketing Mix Models with pymc-marketing. Use when building a new MMM, choosing adstock/saturation transformations, specifying priors with moment matching, configuring the likelihood, setting up the model constructor, or designing the fitting strategy. Also activate when discussing prior calibration, spend-share sigma, channel prior tables, or model_config dictionaries.

Multi-geo panel MMM using pymc-marketing's multidimensional API. Use when the dataset has multiple geographies, DMAs, countries, or regions that should be modeled together. Covers geo column setup, panel data validation, and the multidimensional MMM constructor.

Stakeholder-specific MMM reporting templates and content guidance. Use when generating or reviewing CMO, CFO, Marketing Ops, or Data Science reports, or when explaining what each report should contain and how to frame results for each audience.

Target unit handling for MMM: monetary vs acquisition vs volume targets, CPA vs ROAS framing, and value-per-unit conversions. Use when the target variable is not a currency amount (e.g., policies sold, signups, app installs), when designing the spec.yaml target_unit field, or when a CFO report needs to show both CPA and revenue-equivalent ROAS.

Manually log a bug into the neural memory knowledge graph when not using context log files. Links the bug node to the affected code by file path.

Manually log a task or phase into the neural memory knowledge graph. Links the task to related code files so it appears in semantic search results.

View and modify neural memory settings — indexing mode, exclusions, redaction patterns, and staleness thresholds.

Get a compact, token-budgeted context snapshot AND save a rich persistent session snapshot — project overview, active bugs/tasks with code connections, and recent git commits. Use this to orient quickly and persist context for the next session.

Build the complete neural memory knowledge graph for this codebase by parsing AST, resolving relationships, and scoring importance.

Generate comprehensive technical documentation by synthesizing all accumulated insights from the neural memory knowledge graph. Also use to save a new insight with neural_add_insight.

Deep-dive into a specific code element — see its full context, callers, callees, siblings, and call chains in the knowledge graph.

Search the neural knowledge graph for functions, classes, modules, or concepts and get layered results with summaries.

Check the health and freshness of the neural memory index — initialization state, staleness, and graph statistics.

Stop the running neural memory dashboard server.

Manage the full task lifecycle in the neural knowledge graph — list, create, update status and priority. Tasks connect to code nodes and appear in every prompt via the context hook.

Sync neural memory with recent code changes incrementally without a full re-index.

Generate and open the interactive knowledge graph dashboard in your browser — hierarchy treemap, semantic radial tree, and force-directed graph.