skillopt

Change management, SkillOpt optimization, and evidence-gated evals for agent skills.

Darwin Skill 2.0: autonomous evaluator and optimizer for Agent SKILL.md workflows

Nightly self-improvement for your DeepSeek Harness agent: harvest past sessions, replay recurring tasks, and consolidate validated skills behind a held-out gate. Brings microsoft/SkillOpt's SkillOpt-Sleep engine to dsh as native tools + a bundled skill (MIT).

Bilingual SkillOpt-inspired Codex skill writer for SKILL.md creation, validation gates, bounded edits, and agent skill optimization

AI artifact lifecycle management for prompts and agent skills: optimize, evaluate, version, trace, and promote changes from raw data and policies.

A Codex skill that researches and learns unfamiliar domains before building.

Obsidian + Hermes agent intelligence economy kit: clippings harvest, skill optimization, token efficiency, verification, loop scoring.

A Claude Code skill that researches and learns unfamiliar domains before implementation.

Systematic methodology for optimizing agent skills using deep-learning-style approach. Based on arXiv:2605.23904

Train AI agent skills like model parameters. Offline text-space optimizer implementing SkillOpt (arXiv:2605.23904v2) for Hermes Agent. Bounded edits, validation gating, zero inference overhead.

Self-improving memory for coding agents.

Codex plugin for freebuff2api with a SkillOpt self-improvement loop

EvoSkillOpt — When SkillOpt Meets Evolutionary Game Theory (EGT) and Reinforcement Learning (GRPO)

Train agent skills like neural networks — with epochs, batch size, learning rates, and validation gates — without touching model weights. Based on Microsoft SkillOpt.

Optimize local AI agent workflows and system prompts using an iterative evaluate-and-refine process while keeping your model weights frozen.