openai openai-agents-js
OFFICIALLABSCO SUMMARY
The sixteen skills fall into four dependency classes: ten need nothing beyond the checked-out repo (local_tool), three are ready to run with no setup at all, two need an OpenAI account and API key, and one talks to an MCP service. What they actually do is entirely about the codebase's own upkeep: validating a changeset against a git diff, running the required install/build/type/lint/test/format gate before a change ships, auditing the authored English SDK docs against the implementation, keeping pnpm itself current, drafting a PR title and description, judging whether a release candidate should ship, choosing the supported scope for a behavior change, giving a GitHub issue or PR a desk review for demonstrated need, pulling current OpenAI API documentation as evidence for a claim, and running an approved runtime-behavior probe.
Three of the sixteen don't fit that pattern: credit-note-fixer, invoice-total-fixer, and csv-workbench, described respectively as fixing a formatting bug, fixing invoice-total math, and summarizing CSVs in a sandbox quickstart repository. The README explains the likely reason without naming these three directly: openai-agents-js ships a beta Sandbox Agent that gives an agent a filesystem workspace, and these read as that feature's own worked exercises, packaged as skills alongside the real maintenance tooling.
READ THE FULL ANALYSIS
Who this is for. Not someone who wants to learn to build agents with @openai/agents — none of the sixteen teaches that; the SDK's own guides for agents, tools, guardrails, handoffs, sessions, and tracing live in its documentation, not in this skill set. This package is for someone extending or reviewing changes to openai-agents-js itself, on Node 22+, Deno, or Bun, inside its own pnpm workspace with Changesets and Vitest already set up.
What it costs to start. Two skills need an OpenAI API key and one needs an MCP connection; the other thirteen run against the local git checkout with no external account at all.
ALSO IN THIS PACKAGE
pnpm upgrade
WHAT'S INSIDE
14 showing · 14 totalNothing else to set up — install it and go.
changeset-validation
Every change to this project ships with a short note saying what changed and how big a release it warrants; this checks that the note matches the code and carries the right size.
code-change-verification
The last check before a change is called finished: the whole project is installed, built and tested from scratch, and nothing counts as done until every step comes back clean.
credit-note-fixer
A one-bug repair job in a practice repository: make the credit note print the right label and a positive amount, then run the one test that proves it.
csv-workbench
Point it at a spreadsheet-style data file and get the totals and averages back in a few lines, with any guesses about missing or broken rows spelled out.
docs-sync
Compares this project's English documentation against the code as it stands now, reports what is missing or wrong, and — once you approve — goes and fixes the pages itself.
final-release-review
A go/no-go call on the next release: everything that changed since the last one, the risky parts ranked, and — for a feature release — a draft of the release notes.
implementation-strategy
A way of deciding how far a fix should go before you write it: the smallest thing that must work, what you are deliberately not going to support, and when to stop patching and rethink.
invoice-total-fixer
In a small practice repository, the invoice total adds tax the wrong way; this corrects it to charge tax as a percentage of the subtotal and reruns the tests to prove it.
maintainer-review
Someone files a report or offers a patch — this decides what the owners should do about it, starting from whether a real problem was ever shown, not from whether the code looks fine.
openai-knowledge
Looks up the exact wording of OpenAI's own documentation before answering, so field names, defaults and limits come off the page rather than out of memory.
pnpm-upgrade
pnpm is the tool that installs this project's dependencies; this moves it to the newest release everywhere it is written down — your machine, the project settings, and every automated build.
pr-draft-summary
Writes the title, description and branch name for the change you just finished, ready to paste into GitHub — and stops there, without creating or pushing anything itself.
runtime-behavior-probe
When reading the code cannot settle how something actually behaves, this plans a small throwaway experiment, shows you exactly what it would run, and only runs it once you say yes.
test-coverage-improver
Some of the code is never touched by any test; this works out which parts, writes tests for the ones that matter, and measures again to show the hole is filled.
HOW TO GET IT
npx skills add openai/openai-agents-jsnpx skills add openai/openai-agents-js --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.