Automate creation and optimization of Claude Code skills with progressive disclosure.
npx skills add https://github.com/resolve-io/.prism --skill skill-builder-resolve-io
Please help me install this Agent Skill. Skill: skill-builder Source: https://github.com/resolve-io/.prism/tree/main/skills/skill-builder Command: npx skills add https://github.com/resolve-io/.prism --skill skill-builder-resolve-io
Building efficient, scalable Claude Code skills can be complex, leading to token waste, slow performance, and difficult maintenance. This skill provides a structured approach to overcome these challenges, guiding you through best practices for token optimization and progressive disclosure.
SKILL.md templates, process guides, and validation checklists for planning, structuring, and deploying your skills.Help me create a new Claude Code skill for incident triage. What are best practices for skill token optimization? Show me a SKILL.md template.
High-intent search queries and answers about installing and using this skill.
Structure skills in three progressive levels: metadata (~100 tokens in SKILL.md), body under 5k tokens with core logic, and bundled reference files deferred to Level 3. This three-level loading pattern reduces cognitive load and token waste by disclosing only essential information upfront.
Use ready-to-use SKILL.md templates, process guides, and validation checklists to plan, structure, and deploy skills consistently. Templates ensure metadata stays concise, skill bodies remain focused under 2k tokens, and supporting details live in /reference files for deferred loading.
Follow the three-level loading pattern: write Level 1 metadata in SKILL.md (~100 tokens), develop Level 2 body with core functionality (under 5k tokens), and organize Level 3 bundled reference files. Use provided templates and checklists to guide planning, structuring, and deployment.
Yes. Refine existing skills by redistributing content across the three-level pattern: condense metadata to ~100 tokens, trim body to under 2k focused tokens, and move detailed references to bundled /reference files. This improves performance and reduces token overhead.
Level 1 metadata targets ~100 tokens in SKILL.md, Level 2 body should stay under 5k tokens with core logic concentrated under 2k, and Level 3 bundled reference files hold supporting details. These limits enforce progressive disclosure and prevent token bloat.
open-sessionParse and validate plugin settings from markdown files with YAML frontmatter.
woxiangyangzhimaoAudit token usage in Claude Code sessions and identify redundant components.
terrylicaGenerates and enforces Claude Code CLI skill architecture templates and validation scripts.
haxlysArchitect backend systems with Clean Architecture, Hexagonal Architecture, and Domain-Driven Design.