agents-md-skill

A February 2026 study (arXiv:2602.11988v1, ETH Zurich) across 438 tasks and 4 coding agents found:

Analyze the current conversation to extract the key workflow steps, agentic orchestration patterns, and decisions. Transform these into a reusable slash command that captures BOTH what was done and how it was orchestrated.

Create or improve AGENTS.md, CLAUDE.md, CODEX.md, or similar repository-level context files for coding agents. Also capture conversation workflows into reusable slash commands. Use this skill whenever the user wants to create, write, edit, update, or optimize an AGENTS.md or CLAUDE.md file, or asks about how to guide AI coding agents in their repository. Also trigger when users mention "context file for agents", "agent instructions", "coding agent setup", repository onboarding for AI, or want to improve how Claude Code, Codex, Qwen Code, or other coding agents work with their codebase. Trigger for "capture workflow", "save this as a command", "create a slash command from this conversation", or similar requests to extract reusable patterns from conversations. This skill encodes empirical research findings on what actually helps vs. hurts agent performance — most auto-generated context files actively decrease success rates and increase costs, so getting this right matters. The same principles apply to slash commands: every line costs reasoning tokens. Trigger even for non-Python repos — the principles are language-agnostic.