Write once, reuse across tools

This page is about where a skill can travel. The short version of the previous lessons: a skill is a folder with a SKILL.md inside (name and description frontmatter, the procedure as markdown), discovered in project locations like .github/skills, .claude/skills, or .agents/skills, and user locations like ~/.copilot/skills. Those project locations reflect a shared Agent Skills format. The SKILL.md you package is designed to be portable - but portability is still subject to each tool’s version, supported locations, and which feature surfaces (chat, agent runs, cloud agents) load skills at all.

One file, supported surfaces

The audit-event skill (the team’s event-recording procedure, packaged in the previous lessons) can be reused across Copilot surfaces that support Agent Skills:

• VS Code - where you worked this whole chapter: Copilot pulls the skill in when a task matches the description.
• The Copilot CLI - current versions support the shared skill locations; verify the version and its discovery output.
• The cloud coding agent - repository skills may be available to the agent, subject to the cloud surface and repository policy (check the customization support matrix and your repository’s policy).

So the procedure you wrote once travels with the work - place it in a project location and it travels with the repo, whoever - or whatever - picks it up.

The convergence payoff

Now the part that reaches past Copilot. Back in the Rules chapter you saw the first convergence beat: the AGENTS.md and CLAUDE.md files written for another agent were read natively by Copilot, no translation. The same context, honored by a different tool.

Skills are the second beat of the same story. The audit-event skill from this chapter follows the same format used in the Claude Code course and the Codex course. A skill authored there may work here unchanged - check the Agent Skills documentation and each tool’s supported locations before treating that as a guarantee, just as with rules files.

That pattern arriving twice is not a coincidence. The hard part of working with agents is not the tool - it is closing the gap between what the agent knows and what your team knows: your conventions, your procedures, the audit-event shape everyone has in their head. You close that gap by writing it down once. Rules proved your always-on context survives the jump between tools; skills prove your on-demand procedures do too. You are not re-teaching the audit convention to each new agent you adopt - you taught it once, in a standard format, and every conforming tool already speaks it.

Where this leaves the work

The audit-event convention is now a portable skill Copilot invokes on its own, in orders-service (the course’s illustrative order service) and shared-lib (the internal library it imports) alike, in a format your other agents may already understand. The procedure is handled. What is still missing is reach: the skill knows how to record an event, but it cannot see the internal systems the work depends on - the issue tracker, the service registry that lists which consumers a shared-library change touches. Giving Copilot a live line into a system outside the repo means wiring an MCP server. Next chapter: extending Copilot over MCP.