joneslloyd/agent-skills

This repository is a collection of skills designed to extend AI coding assistants such as Claude Code. Skills in this context are small, structured instruction sets that tell the AI how to perform a specific task when you invoke a command. The repository also serves as a Claude Code plugin marketplace, meaning you can install its skills directly inside Claude Code with a single command rather than copying files by hand. The first and currently only skill included is called counterpart-analysis. You give it a transcript of a conversation, an email thread, a chat log, or a description of a situation involving other people, and it tries to work out what each person actually wants below the surface. Rather than summarizing what was said, it focuses on underlying interests, incentives, and constraints that may not be stated openly. The skill builds its analysis carefully. Every claim about a person's motives is tied back to a specific quote or observable pattern in the material. Each piece of information is labeled to show whether it came directly from the text, from background you supplied, from the skill's own reasoning, or from an untested assumption. The skill also generates two or three competing explanations for the same behavior, ranks them, and suggests a concrete question or test to figure out which one is correct. It includes guardrails to avoid over-confident readings, especially when the material is thin or the conversation involves cultural or language differences. The skill is designed for understanding situations and preparing for conversations, not for writing manipulation scripts. An explicit ethics boundary is part of the design. Installation works through Claude Code's plugin system, through OpenCode by copying the skill folder, or for claude.ai users by downloading a packaged file and uploading it in a conversation. The repository is MIT licensed, meaning you can use, copy, and modify it freely.