Doing it well
Use human gates, safe configuration, focused context, evidence, usage signals, and rollback points throughout either project.
Open these pages when a project needs the practice. They are not an entrance exam.
| Practice | Use it when |
|---|---|
| Approval gates | An action changes code, data, external systems, or production |
| Secrets and environment variables | A project needs credentials or per-environment configuration |
| Context engineering | An agent is guessing, wandering, or reading too much |
| Evals and evidence | “It seems to work” is not enough |
| Cost and observability | A feature consumes model or third-party resources |
| Stop, rollback, and maintain | A change should ship, pause, or be reversed |
Approval gates
Place a specific human decision before code, data, external, destructive, and production actions instead of reviewing after the consequence.
lessonSecrets and environment variables
Classify configuration by exposure and place real values safely in local, preview, and production environments without committing them.
lessonContext engineering
Assemble the smallest authoritative working set—task, sources, state, constraints, tools, examples, and stop conditions—for an agent.
lessonEvals and evidence
Turn requirements into a fixed set of repeatable checks so code, prompt, and model changes can be compared rather than admired.
lessonCost and observability
Record the few usage, latency, failure, and capacity signals that answer an operational question without installing every dashboard.
lessonStop, rollback, and maintain
Use scope boundaries, diffs, checks, health evidence, and save points to decide when to ship, pause, reverse, or update.