Super Skills for Claude Code

This file provides the Claude Code view of the unified skill repository.

Purpose

This repository keeps workflow logic shared while adapting packaging per host.

• Shared authored skill source lives in skills/
• Claude-ready skill payloads live in .claude/skills/
• Codex-ready skill payloads live in .agents/skills/

Claude Skill Layout

Claude Code should consume skills from:

.claude/skills/<skill-name>/SKILL.md

Those skill directories are generated from the shared skills/ source tree. Do not edit generated Claude skill copies directly.

Source skills stay host-neutral. Claude-only command, hook, and agent metadata should not be authored in skills/.

Current Skill Coverage

Core workflow skills:

• plan-product
• plan-architecture
• investigate
• review
• security-review
• qa-browser
• ship-release
• health-check
• checkpoint
• careful
• freeze
• guard

Knowledge-pack skills are installed alongside the core set unless profile selection narrows them.

Security Stance

• Keep hooks opt-in
• Do not assume telemetry
• Do not auto-approve actions
• Keep destructive actions explicit
• Prefer environment variables over plaintext secret files

Host Boundary

Claude-specific adapter logic should stay thin. Workflow semantics belong in shared skills, manifests, and docs, not in hidden host glue.

Claude-only best-practice material can inform:

• adapter templates
• runtime guidance docs
• host-specific install steps

It should not redefine the shared source skill contract.