jonathanong/jong-skills
Generic agent skills that explain a design or a procedure: CI and dependencies, PostgreSQL and queues, testing, TypeScript, Rust, React and native apps, HTTP APIs, rollout, logging, pull requests, docs, and evals.
Carry a repository change from start to handoff - implement it, gather evidence, review a diff, and answer review feedback, splitting independent work to run in parallel. Use when starting an implementation task, preparing a handoff, reviewing a change, or deciding what to do with review comments from a person or a bot.
Bound the rows each run reads when writing or reviewing scheduled jobs, reconcilers, cleanups, dispatchers, backfills, exports, list endpoints, or any loop that reads rows.
Configure Dependabot coverage, grouping, cooldowns, and secrets, and start a Dependabot update check on demand. Use when writing or reviewing dependabot.yml or a workflow that runs on Dependabot pull requests, or when asked to trigger, re-run, or force a Dependabot run.
Decide what belongs in docs, code comments, and agent instruction files, so they do not drift from the code. Use when writing or reviewing a README, a docs page, a code comment, or an AGENTS.md / CLAUDE.md file, including a diff that adds or changes any of them.
Lay out and maintain an eval suite's documents - README.md for current scores and scope, HISTORY.md for the run log, AGENTS.md for the rules. Use when creating an evals directory, adding or changing an eval case, grader, or fixture, recording or reporting eval results, or reviewing a change that quotes an eval score.
Ship new behavior behind a flag that starts off, treat deploying and enabling as separate steps, and add no table or service that no entry point reaches. Use when adding or reviewing a feature flag, a flag's default, rollout or rollback steps, or a change that adds a table, migration, service, or write path.
Check scope, staged content, and message before a commit, and rebase or force-push a shared branch without overwriting someone else's work. Use before every git commit, before any rebase, reset, or force push of a branch that another person or agent may also push to, and when a rebase or merge conflicts in a lockfile.
Design GitHub Actions workflows that react to events instead of polling, and check triggers, concurrency, timeouts, runners, and pins. Use when writing, editing, or reviewing a workflow or composite action.
Search for, write, label, link, and close GitHub issues without filing duplicates or writing where you may not. Use when filing or editing an issue, recording a follow-up, closing an issue, or linking a pull request to an issue.
Restrict who can open pull requests on a GitHub repository and gate outside contributors' workflow runs, with the REST calls to read, set, and roll back each setting. Use when hardening a public repository, limiting pull requests to collaborators, or deciding when fork pull requests may run Actions.
Design glide-mq queues, jobs, ordering lanes, worker concurrency, rate limits, and cross-job caps. Use when writing or reviewing a queue producer, worker, or dispatcher, choosing which queue or lane a job belongs in, or handling a request that waits on a job's result.
Shape an HTTP API around resources, thin route handlers, user-facing identifiers, and one owner each for secret storage and cache headers. Use when writing or reviewing an endpoint, route handler, or service layer, choosing identifiers in URLs, storing tokens or one-time codes, or setting cache headers.
Structure a native app (Swift, Kotlin, .NET MAUI) around a portable core library with thin platform views. Use when writing or reviewing native app code, deciding where logic lives, adding a web view, setting up native project files, or planning native CI.
Prepare a first npm publish and trusted publishing (OIDC) for a human to run, and set up the GitHub Actions job that publishes. Use when publishing a new package, moving a package to trusted publishing, or fixing a CI publish that fails with E404 or ENEEDAUTH.
Check package metadata, dependencies, entrypoints, and lockfiles.
Plan a repository change - compare the options, trace what the change touches, order the steps, and save the plan where it will be found. Use when asked for a plan, a design, or an approach before implementing, or when a task is large or uncertain enough to need one.
Decide what a browser test should cover and how to locate elements with roles, labels, and test ids. Use when writing, reviewing, or de-flaking Playwright or other browser end-to-end tests, or when adding test ids to interface code.
Write, measure, and improve PostgreSQL queries in Node.js services - query shape, indexes, EXPLAIN plans, round trips, transactions, and pooling. Use when writing or reviewing a SQL query or data-access function, adding a query with a new filter or sort, reading a query plan, or fixing a slow query.
Design PostgreSQL partitioning and partition pruning around UUIDv7 keys. Use when writing or reviewing a partitioned table, a retention or partition migration, a query against a partitioned table, or a decision on whether to partition.
Design PostgreSQL tables, names, relations, history, work-item and cursor tables, retention, and locking. Use when writing or reviewing a migration, a table, column, or foreign key, or any code that takes a row lock (FOR UPDATE) or an advisory lock, serializes concurrent writers, or enforces a cap, limit, or last-one guard under concurrency.
Write a pull request description a reviewer can act on without opening the diff - the problem, the change, the impact, and before/after evidence. Use when opening or updating a pull request, writing its title or body, deciding whether to split work across pull requests, or reviewing a description.
Structure a React or Next.js app so its API calls, links, redirects, and configuration each have one traceable owner. Use when writing or reviewing React components, Next.js pages, layouts, or route handlers, data fetching in pages, auth redirects, entity links, or browser-side configuration.
Diagnose a CI failure or noisy CI output from the logs, find the first cause the repository owns, and fix it without hiding diagnostics. Use when a CI run fails or flakes, when asked to make CI green or quieter, or when asked to add a retry or silence output in a workflow.
Find deferred work that was promised in merged pull requests, closed issues, and instruction files, check that it is still needed, and draft issues for it. Use when asked to sweep for follow-ups, unfinished checklists, or dropped work over a recent period.
Lay out Rust tests and caches, and run heavy cargo builds without exhausting a shared machine. Use when writing or reviewing Rust code or tests, adding a cache, or running cargo build, test, or coverage jobs, especially from several agents or worktrees at once.
Keep the boundary between services, and between a service and an outside partner, explicit - own your data, map foreign payloads once, keep raw bodies for replay, and test the contract. Use when writing or reviewing code that calls another service or a third-party API, handles a webhook, reads another service's data, or puts deployment topology in a repository.
Place, author, and validate one static-analysis invariant. Use when adding or changing a lint rule, CI guard, or policy check that blocks a pattern in code, configuration, or workflows.
Choose and wire up static code analysis for a repository - which linters, formatters, type checks, dead-code, dependency, workflow, secret, and infrastructure checks to add, with the settings to start from. Use when creating a new repository, adding a language or toolchain to one, or auditing which static checks a repository is missing.
Write log lines with a static message and structured fields, and load their context without extra queries. Use when adding or reviewing log statements, logger calls, or the data a log line carries.
Decide which tests to write and how to build fixtures, mocks, helpers, and guards so each test fails for one real reason. Use when writing, reviewing, fixing, or deleting any automated test, fixture, or test helper - Vitest or Jest, backend integration, Next.js, Storybook stories, Swift, or .NET. For browser tests, also use playwright-authoring.
Triage Codex Security findings for one repository into evidence-backed, explicitly approved outcomes. Use when clearing or reviewing a Codex Security finding queue; leave repository issue management to the consumer.
Keep TypeScript and Node.js code direct - no forwarding aliases, shims, or dynamic dispatch - with errors handled where thrown, one batch call for per-item I/O, and packages that respect layer and runtime boundaries. Use when writing or reviewing TypeScript or Node.js code, especially renames, async errors, loops that query or enqueue, bulk paths, and HTTP calls.