| name | cross-harness-skill-authoring |
| description | Use when creating, reviewing, or adapting an agent skill that should remain portable across Codex, Claude Code, and other compatible harnesses. |
Cross-Harness Skill Authoring
Use this skill when a skill should travel between agent harnesses without depending on one runtime's private assumptions.
Workflow
- Treat
SKILL.md as the canonical portable behavior.
- Keep the skill directory name stable, lowercase, and kebab-case.
- Put harness-specific launch, install, validation, or tool notes in separate sections instead of mixing them into the core workflow.
- Use paths relative to the plugin or skill directory. Installed plugins are commonly copied into cache directories, so references outside the plugin may break after installation.
- Prefer instructions that describe intent, inputs, outputs, and verification over instructions tied to one chat UI or model provider.
- When a harness requires metadata, add only the fields that harness accepts and keep the meaning aligned with the canonical skill.
Compatibility Checklist
- The skill lives at
skills/<skill-name>/SKILL.md.
- Frontmatter includes a concise
description.
- Optional frontmatter fields are additive and do not change the skill's meaning for another harness.
- External assets, scripts, and references live inside the plugin directory.
- Codex metadata points
skills at ./skills/ in .codex-plugin/plugin.json.
- Claude Code metadata points
skills at ./skills/ in .claude-plugin/plugin.json.
- Marketplace entries point at the plugin directory with a relative
./plugins/<plugin-name> source.
Validation
For Codex packaging, run the Codex plugin validator against the plugin directory.
For Claude Code packaging, run claude plugin validate . at the marketplace root and against the plugin directory when the Claude CLI is available.
