Objective
Translate harness principles into enforceable implementation and operational patterns.
Use This Skill When
- •Designing new agent context files, templates, or lifecycle workflows.
- •Evaluating whether a solution relies on trust vs mechanical enforcement.
- •Improving error messaging, remediation loops, or drift visibility.
Procedure
- •Keep context short, structured, and pointer-first.
- •Encode conventions in templates, rules, tests, or linters.
- •Build feedback loops that let agents self-correct from observable state.
- •Add status outputs that make drift visible without manual inspection.
- •Prefer reusable constraints that scale across agents and codebases.
Guardrails
- •Avoid monolithic instruction files.
- •Avoid manual synchronization checklists as primary control.
- •Prefer deterministic systems over subjective reminders.
Done Criteria
- •New behavior is mechanically enforced, not advisory.
- •Error outputs include direct remediation guidance.
- •Cross-agent drift risk is reduced by design.