Use One Skill Set Across Claude Code and Other Agents

Since Agent Skills was published as an open standard (December 2025, at agentskills.io), the same SKILL.md format is read by a growing list of agent products: Claude Code and claude.ai, plus tools from other vendors including OpenAI Codex, GitHub Copilot, Cursor, and more. Practically, that means a skill is no longer tied to the tool it was written in. Here is how to actually get that portability.

Keep one canonical skills repo

Maintain your skills in a single version-controlled folder: one directory per skill, SKILL.md inside each. This repo is the source of truth; every tool gets its skills from here, never by hand-editing an installed copy.

my-skills/
  release-line/SKILL.md
  bug-triage/SKILL.md
  deploy-checklist/
    SKILL.md
    scripts/verify.py

Write for portability

• Reference bundled files by relative path only, never a machine-specific absolute path.
• State prerequisites explicitly in the body (tools, credentials, services) so a new environment can meet them instead of guessing.
• Keep bundled scripts self-contained (standard library only, no install step) so they run wherever the skill lands.
• Do not lean on one tool’s private features in the core instructions; if a step is tool-specific, label it as such so other agents can skip it gracefully.

Install per tool, verify per tool

Each agent has its own skills location and install mechanism: a directory to copy into, a plugin system, or a config pointer. The portable part is the skill itself; the wiring is per-tool. After wiring each tool, run the same two-prompt check: one request that should trigger the skill, one near-miss that should not.