skill-opt

name	skill-opt
description	Use to optimize/improve an existing agent skill against scored tasks. Ports Microsoft SkillOpt: treats the target SKILL.md as a trainable document and improves it via a rollout→reflect→edit→gate→memory loop with a held-out gate, while keeping the model frozen. Trigger when the user wants to tune, harden, or measurably improve a skill.

skill-opt

What This Does

One agent switches hats: Setup → Rollout → Score → Reflect → Edit → Gate → Memory, then repeats. The run directory .skill-opt/runs/<skill>-<n>/ is the message bus — every phase reads/writes files there, making runs resumable from any point. The one-man-play is the backbone, not a constraint: when parallelism > 1, rollouts and gating fan out to fresh subagents.

Up-Front Questionnaire

Ask the user these questions, then write config.yml from templates/config.yml:

Knob	Default	Notes
target_skill	— (required)	path to SKILL.md or skill dir
edit_references	false	also edit references/?
feedback_source	proposed-ratified	proposed-ratified | autonomous | user-suite | live
feedback_timing	autonomous	autonomous | interactive (pause at each gate)