OpenAI Plugins: package examples, portable formats and connection boundaries
Explore OpenAI Plugins safely: inspect one package before installing anything
Use the Figma example to trace manifest paths and connection assumptions, then define an isolated skills-only trial with a clear acceptance condition.
What you will learn
- Start with a file inspection, not a bulk installation
- Use a small local learning workflow first
- Decide success from the actual output
Before you start
- Basic JSON and directory-path knowledge
- Understanding of skills and external-service permissions
Distinguish source examples from current format guidance and inspect a package without confusing metadata with runtime proof.
Key takeaways
- Inspect one package and its referenced resources first.
- Use synthetic skills-only work to test the basic workflow.
- Discovery and correct output require separate acceptance checks.
Start with a file inspection, not a bulk installation
Choose one package instead of installing the entire catalogue. In the Figma example, read the manifest, the referenced skills directory and .app.json. The manifest’s apps path points to the package root, not into .codex-plugin. Follow the actual directory tree when documentation labels and on-disk skill names differ.
The inspection worksheet below is illustrative data, not a command that changes your profile. Record package revision, declared version, referenced resources and the external service it expects. A missing path is a packaging problem; an existing path does not establish that its contents are safe or compatible with your host.
Use a small local learning workflow first
For a first authoring trial, a skills-only package avoids external-service authentication. Define a synthetic meeting-note workflow that extracts decisions, owners and open questions without sending messages. Current official guidance recommends a portable root manifest plus a skill under skills. This suggested exercise does not install or create a package on your machine.
If you choose to install later, use the host’s documented local marketplace process and a new conversation for the trial. Preserve existing marketplace entries and inspect the installed source and version. Do not paste example configuration over an existing personal catalogue or connect a real service merely to test basic discovery.
Decide success from the actual output
Give the trial invented notes containing one explicit decision, one named owner and one unresolved question. Check that it preserves each distinction and does not invent deadlines or send a follow-up. Package discovery and correct workflow output are separate acceptance checks; passing the first is not enough.
Then evaluate a connected example only if the task needs it. Confirm authentication and permitted read/write scope with synthetic or disposable resources before real documents. We did not run this host trial, so the proposed exercise remains a reproducible plan rather than a measured plugin success rate.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Trace manifest paths from the plugin root.
- 2
Define an invented task and prohibited side effects.
- 3
Review local installation scope before changing configuration.
- 4
Test in a new conversation and inspect the output.
Copy-ready example
{
"inspectionOnly": true,
"package": "figma",
"manifest": ".codex-plugin/plugin.json",
"skills": "./skills/",
"apps": "./.app.json",
"profileChanged": false
}Frequently asked questions
Should I install every catalogue entry to learn the format?
No. One small example is sufficient to inspect the packaging boundaries.
Does this article connect my Figma account?
No. It describes an inspection and trial plan without installing or authenticating anything.
Sources
- OpenAI Plugins / README.mdSource checked 2026-09-14
- OpenAI Plugins / plugins/figma/.codex-plugin/plugin.jsonSource checked 2026-09-14
- OpenAI Plugins / plugins/figma/.app.jsonSource checked 2026-09-14
- OpenAI — Package your plugin (checked 2026-09-14)Source checked 2026-09-14