progressive-disclosure

Audit Agent Skill context overhead in OpenAI Codex and Claude Code; reversibly lazy-load approved Codex skills without changing triggers, with safe rollback.

LUCID: reader-sovereign progressive disclosure. Mark a term once; open it at three depths, gloss to explainer to detail with a path to source. Framework-agnostic core plus a React binding.

Agent Skills 学习工坊 · 基于 agentskills.io 官方规范整理的 55 个核心知识点（SKILL.md、YAML frontmatter、渐进式披露、最佳实践、Evals、客户端接入、生态与安全）

A reusable Company OS for Claude Code, Codex, progressive disclosure, durable ownership, shared skills, and connected repositories.

Human-paced AI responses with progressive disclosure and transparent usage tradeoffs.

Search a separate local Agent Skills library by name or task and load only the selected SKILL.md.