skillopt-sleep

Harvest transcripts, mine patterns, and test SKILL.md edits against a held-out gate.

15.8k|1.5k|Updated May 8, 2026
One-click install
npx skills add https://github.com/microsoft/SkillOpt --skill skillopt-sleep
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: skillopt-sleep
Source: https://github.com/microsoft/SkillOpt/tree/main/plugins/openclaw
Command: npx skills add https://github.com/microsoft/SkillOpt --skill skillopt-sleep

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Validating and refining agent skills through nightly sleep cycles with held-out gates, enabling safer, gated self-improvement of skills.

Core Features & Use Cases

  • Nightly harvesting of session transcripts and mining recurring patterns.
  • Testing proposed skill edits against a held-out validation score before staging.
  • Human-in-the-loop adoption to ensure safety and control.

Quick Start

Run a full nightly sleep cycle with the default config to start refining your skill with gated proposals.

Frequently Asked Questions about skillopt-sleep

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate nightly self-improvement for AI agent skills?

You automate nightly self-improvement for agent skills by harvesting session transcripts, mining recurring patterns, and replaying tests with baseline and candidate edits. This loop stages improvements only when they outperform the baseline.

What is a held-out validation gate for testing skill edits?

A held-out validation gate is a safety mechanism that tests proposed skill edits against a held-out validation score. It ensures only improvements that outperform the baseline are staged for adoption, preventing regressions.

How do I validate proposed agent skill edits before adoption?

You validate proposed agent skill edits by replaying tests with baseline and candidate SKILL.md edits during a nightly sleep cycle. A held-out gate compares performance to ensure only safe improvements are staged for human-in-the-loop adoption.

Do I need a running engine and test sets to use nightly skill refinement?

Yes, nightly skill refinement requires a running SkillOpt-Sleep engine, a compatible backend, and access to test sets. These components are necessary to safely operate the gated self-improvement cycle and validate candidate edits.

How does human-in-the-loop adoption work for evolving agent skills?

Human-in-the-loop adoption for evolving agent skills works by staging validated improvements that pass a held-out gate. The system automatically mines patterns and proposes edits, but a human reviews and approves the staged changes for safety.

What are the limitations of using a nightly self-improvement loop for skills?

A limitation of the nightly self-improvement loop is its strict requirement for a running engine, compatible backend, and accessible test sets. Without these, the system cannot safely replay tests or apply the held-out gate to validate improvements.