Turning 15 Skills into an Agent You Can Verify
Organize 15 Skills with task protocols, work and acceptance chains, six failure tests, and a practical path from one frequent task to a reliable Agent.
Knowledge path
OpenAI
What Is Published Is a Snapshot, Not a Real-Time System
Deconstructing Publish Artifact to Sites: prerequisites for publishing, the one-to-one mapping between logical artifacts and sites, the distinction between access policies and editing permissions, and why automated publishing is conditional.
Dashboards and Reports Share Evidence, but Serve Different Jobs
Compare Build Dashboard and Build Report: their audiences, metric models, evidence requirements, acceptance criteria, and delivery conditions.
Reproducible Calculations Make Charts Credible
How Jupyter Notebooks and Visualize Data connect reproducible calculations, chart contracts, and honest presentation in a Data Agent workflow.
Data Quality and Validation: Two Gates, Two Risks
Separate checks on source data from independent checks on analytical claims, including grain, missingness, joins, denominators, and causal interpretation.
How Big Is the Market? Deliver an Estimate You Can Interrogate
How Market Sizing fixes boundaries and units, separates facts from proxies and assumptions, and uses sensitivity to pick what to verify first.
When Metrics Drop: Reproduce the Change Before Discussing Causes and Actions
How Metric Diagnostics reproduces a change before explaining it, splits drivers by numerator, denominator, and group, then scopes the cause.
From Defining Success to Sustained Operations: The Complete KPI System Loop
How Design KPIs filters candidate metrics, layers outcomes, drivers, and guardrails, and which six definitions to lock before reporting.