Generic agent skills that explain a design or a procedure. A skill here may say which approach to
pick and why. It holds no organization-specific policy; a consuming repository adds its own rules
on top, in its AGENTS.md.
The repository is one plugin, skills, for both Claude Code and Codex.
| Skill | Use it when |
|---|---|
static-code-analysis |
Creating a repository: which static checks to add and how strict to start. |
static-analysis-checklist |
Adding or changing one static-analysis invariant. |
github-actions-authoring |
Writing workflows: events not polling, triggers, timeouts, runners, pins. |
dependabot |
Configuring Dependabot coverage, grouping, cooldowns, secrets, and runs. |
github-repository-settings |
Limiting pull requests to collaborators and gating outside contributors' CI. |
package-json-checklist |
Changing package.json, a lockfile, or a package entry point. |
npm-publishing |
Preparing a first npm publish, trusted publishing, and the release workflow. |
triage-codex-security |
Triaging Codex Security findings into explicitly approved actions. |
postgres-schema-design |
Designing tables, relations, history, retention, and row locking. |
postgres-node-performance-tuning |
Writing or tuning a query: shape, indexes, plans, round trips. |
postgres-partitioning-uuid-v7 |
Partitioning and pruning a table keyed by UUIDv7. |
bounded-iteration |
Writing a sweep, backfill, dispatcher, or list read that must stay bounded. |
glide-mq |
Designing glide-mq queues, lanes, concurrency, rate limits, and leases. |
evals |
Documenting an eval suite: current scores, run history, and rules. |
pr-description |
Writing a pull request description: problem, change, impact, before/after. |
documentation |
Deciding what goes in docs, comments, and agent instruction files. |
test-authoring |
Which tests to write; fixtures, mocks, guards; Vitest, Storybook, Swift, .NET. |
playwright-authoring |
Writing browser tests: what needs a browser, locators, and test ids. |
typescript-authoring |
Writing TypeScript without aliases, shims, or hidden calls. |
react-app-structure |
Structuring API calls, links, redirects, and config in a React app. |
native-app-architecture |
Structuring a native app around a portable core with thin views. |
rust-authoring |
Laying out Rust tests and caches, and running heavy cargo builds. |
http-api-design |
Shaping an HTTP API: resources, thin handlers, identifiers, secrets. |
service-boundaries |
Calling another service or a partner API, or handling a webhook. |
feature-rollout |
Adding a feature flag, or a table or service that needs an entry point. |
structured-logging |
Writing log lines with a static message and structured fields. |
agent-workflow |
Implementing, handing off, reviewing, and answering review comments. |
planning |
Planning a change: options, what it touches, steps, where the plan lives. |
git-commit-checklist |
Committing, and rebasing or force-pushing a branch others also push to. |
github-issue |
Filing, editing, closing, and linking GitHub issues. |
review-ci-logs |
Diagnosing a CI failure or noisy CI output from the logs. |
revisit-followups |
Sweeping merged work for deferred follow-ups that are still needed. |
Claude Code:
claude plugin marketplace add jonathanong/skills
claude plugin install skills@jonathanong-skillsInside a session, /plugin install skills --marketplace jonathanong/skills does both steps.
Codex:
codex plugin marketplace add jonathanong/skills --ref main
codex plugin add skills@jonathanong-skillsThe manifests set no version, so there are no releases to track. Claude Code treats each commit
on main as an update. Codex picks up new commits when you run
codex plugin marketplace upgrade jonathanong-skills.
.claude-plugin/ Claude Code plugin and marketplace manifests
.codex-plugin/ Codex plugin manifest
.agents/plugins/ Codex marketplace manifest
plugin.json Open plugin manifest
skills/<skill>/ SKILL.md and its references
evals/<skill>/ Eval cases for that skill, one directory per case
evals/README.md Current eval scores and what each case checks
AGENTS.md Rules for contributors and agents, including the skill size limit
To add a skill, create skills/<skill>/SKILL.md and at least one case under evals/<skill>/ that
asserts the skill fired. Keep SKILL.md to 150 lines and put detail in references/.
pnpm run evals:check enforces both, and CI runs it. See AGENTS.md.
Every skill has eval cases run with claude plugin eval.
The current scores and what each case checks are in evals/README.md. The run
log is in evals/HISTORY.md, and the rules for cases are in
evals/AGENTS.md.
pnpm run evals # every skill
pnpm run evals npm-publishing # one skillpnpm install
pnpm run lint
pnpm run format:check
pnpm run evals:checkOther plugins by the same author. Each is installed on its own.
| Plugin | What it does |
|---|---|
| no-mistakes | Answers how code connects from the import graph: what uses a symbol, which tests to run. |
| pr-shepherd | Takes a pull request through CI, review comments, and merge. |
| agent-blackboard | Records agent telemetry so agents can improve from their own sessions. |
| service-status | Tells an agent when a failure is a third-party outage and not its bug. |
| hook-harness | Turns GitHub events into agent runs. |