Automate nightly skill optimization cycles with gated held-out scoring.
npx skills add https://github.com/jiangmin8/skillopt --skill skillopt-sleep-jiangmin8
Please help me install this Agent Skill. Skill: skillopt-sleep Source: https://github.com/jiangmin8/skillopt/tree/main/plugins/openclaw Command: npx skills add https://github.com/jiangmin8/skillopt --skill skillopt-sleep-jiangmin8
Validates and refines agent skills through nightly sleep cycles with held-out gates, enabling safer, progressive improvements without impacting live behavior.
Run one nightly sleep cycle using the provided scripts to harvest transcripts, mine patterns, replay tasks, gate improvements, and stage proposals for review.
High-intent search queries and answers about installing and using this skill.
Automated nightly self-improvement runs a gated optimization cycle that replays baseline and candidate edits, validating against held-out scores to ensure progressive refinement without regressions. Accepted proposals are staged for human adoption while preserving live skill backups.
Running a gated skill refinement sleep cycle requires the OpenClaw and SkillOpt-Sleep stack, a configured DeepSeek API key, and local staging configurations including config.json and run_sleep.py to execute the full harvesting and replay workflow.
Held-out gating prevents regressions by replaying baseline and candidate skills against held-out scores. Candidate proposals must pass these gates before being staged for human adoption, ensuring only validated improvements progress to live deployment.
Yes, the SkillOpt-Sleep workflow automates harvesting session transcripts and mining recurring task patterns within OpenClaw environments. It extracts historical interactions to identify frequent tasks for baseline and candidate skill replays.
The best way to stage candidate skill proposals safely is through a gated optimization cycle that preserves backups of current live skills. This mechanism stages only proposals that pass held-out score gates for human review.
Nightly skill optimization fails without proper local staging configurations. You need the complete OpenClaw and SkillOpt-Sleep stack, a valid DeepSeek API key, and properly configured config.json and run_sleep.py files to execute the cycle.
microsoftHarvest transcripts, mine patterns, and test SKILL.md edits against a held-out gate.
IsNoobgrammerOptimize prompts and scoring configurations through an autonomous generate-evaluate-keep loop.
mehdicMine historical project data to identify patterns and generate predictions.
qcmuuEvolve AI agents across prompts, skills, and memory using LLM-guided mutation cycles.
