yorph-data-analyst

Use this agent when the orchestrator has an approved architecture plan and needs to execute the technical data pipeline. Handles sampling, code generation, validation, and full-scale execution autonomously, then returns a structured result summary. <example> Context: The orchestrator has connected to a data source, profiled it, and the user has approved a transformation plan. user: "Looks good, go ahead and run the analysis." assistant: "I'll use the pipeline-builder agent to execute the approved plan — sampling the data, building the transformations, validating the output, and scaling to the full dataset." <commentary> The orchestrator delegates all technical pipeline work to the pipeline-builder agent. This keeps the user-facing conversation clean while the heavy data engineering runs autonomously. </commentary> </example> <example> Context: The orchestrator needs deeper analytical questions answered from an existing pipeline output during the insights phase. user: "What's driving the drop in Q3?" assistant: "I'll delegate a follow-up query to the pipeline-builder to break down Q3 by segment and identify the largest contributors to the decline." <commentary> The pipeline-builder handles iterative analytical queries delegated by the orchestrator during the insights refinement loop. </commentary> </example>

Load this skill when the analysis involves a waterfall, bridge, or walk chart — for financial attribution, variance decomposition, or any visualization where sequential deltas explain the change between two totals. Contains the required data shape (label, value, bar_type, sort_order), floating bar computation, Chart.js rendering pattern, closure validation, and common pitfalls.

Produce a self-contained HTML dashboard using Chart.js that visualizes the analysis findings. Load this skill every time insights have been produced — it is a required part of the delivery phase alongside insights and trust-report. Also load it when the user asks to "show me a chart", "build a dashboard", "visualize this", "plot this", "graph this", "chart the results", or any request for visual representation of data. Every chart should support a named finding from the insights — no chart exists without a reason.

Load this skill when the analysis involves cohort retention, cohort revenue, or any time-based cohort visualization. Contains the required output tables (long-format heatmap + absolute-time stacked bar), canvas-based rendering, color scale logic, null cell handling, period definitions, and the contractual vs non-contractual distinction.

Guide the user through connecting to their data source and take an initial glimpse. Load this skill at the start of every data analysis session, or whenever the user mentions a data source — files, databases, cloud storage, spreadsheets, anything. Triggers include "connect to my database", "I have a CSV", "here's my file", "connect to Postgres / Snowflake / BigQuery / MySQL", "I want to upload data", "my data is in S3 / GCS", "look at this spreadsheet", "I uploaded a file", or any mention of a data file or database connection.

Produce high-impact, plain-English headline findings from pipeline output. Load this skill every time the pipeline-builder returns a validated result — it is a required part of the delivery phase, not optional. Also load it when the user asks "what does this tell me", "what are the insights", "give me the highlights", "summarize the results", or any variation of "what did we find" resulting in fresh data outputs. The insights skill drives follow-up analytical queries back to the pipeline-builder to go deeper before synthesizing findings.