Skill
Preceptor
Keep agent guidance that earns its place.
What it’s for.
Turn recurring corrections into small cues and compare the same task with and without them. Keep inconclusive results visible and make removal reversible.
Who it’s for
People evaluating and improving recurring agent behavior.
A possible output
A fictional cue comparison recording treatment, baseline, errors and a reversible keep-or-remove decision.
Illustrative example, not a measured result.
Before you begin
A minimum sample count or passing gate is not independent proof of efficacy or generalization.
One part of the work
4 · Make it better
