Skill Claude CodeCodex
Implement observability using the Wide Events (Canonical Log Lines) pattern. Instead of scattering logs throughout a request, accumulate high-cardinality context and emit a single, highly-dimensional structured event per service boundary.
Skill Claude CodeCodex
Write integration tests for Cloudflare Workers using vitest-pool-workers and cloudflare:test. Tests the full HTTP cycle — request through handler, middleware, bindings (D1, KV, R2, DO, Hyperdrive), and back. Use when adding test coverage to a Worker, when a new route/endpoint is created, or when /nightshift, /swarm…
Skill Claude CodeCodex
Use mutation testing to find low-signal unit tests (ones that catch no mutant, or only catch mutants other tests already catch) and delete them without losing any bug-catching power. Use when the user says "prune tests", "delete bad tests", "too many unit tests", "tests that restate the code", "low-signal tests"…
Skill Claude CodeCodex
Build a new family "cast party game" in a day: a shared TV screen cast from the grown-up's phone through the OpenGame (OGS) app, plus a grown-up phone and a kid's iPhone/iPad as controllers. Covers the one-page spec, the room architecture (Cloudflare Worker + one Durable Object per room with actor-kit), designing the…
Skill Claude CodeCodex
Progressive self-improvement for an existing game: grade it like a first-party (Nintendo / AAA) studio on an anchored scorecard, then loop — shoot fixed evidence (contact sheet, recorded multi-screen session with real audio, phone strips, perf, audio report) → a FRESH critic subagent scores every row and ranks fixes →…
Skill Claude CodeCodex
Generate game and storybook art with Codex image generation (ChatGPT plan, no per-image cost) or fal.ai (images, character sheets, poses, parallax layers, background removal) and 3D models with Meshy, at build time or through a server, with keys kept out of the client. Includes putting a real child into a story from…
Skill Claude CodeCodex
Run many game teams at once, wide and deep, toward one studio goal: parallel teams each own a different game or a different version of one, share findings on a studio board, and a coordinator consolidates insights, re-steers, prunes and spins up new teams, with Jon and the kids as final judges. Use for "go wide and…
Skill Claude CodeCodex
Build stunning, playable 3D browser games, explorable worlds and interactive 3D explainers with Three.js, with models, textures and sound generated in code, using an art-directed brief, scripted playtests and a screenshot-and-critic loop that iterates every part of the scene to 8/10+. Use for requests to build a new…
Skill Claude CodeCodex
Design before code for non-trivial work: ground in the existing system (how, why), sketch types, signatures and module boundaries with stub bodies, design it at least twice (arena), screen for shallow modules and leaky seams, then implement against the sketch and surface every deviation. Scrap the sketch when the same…
Skill Claude CodeCodex
Run N parallel attempts at the same task, judge them against a rubric, pick the best as the base, and graft the strongest ideas from the others into it. Use for "arena this", design or concept rounds, a hill-climb that has plateaued, or any non-trivial artifact where one attempt would lock in the wrong shape (an…
Skill Claude CodeCodex
A harsh maintainability review that hunts for restructurings that delete complexity instead of moving it: special cases bolted into unrelated flows, files growing past 1k lines, shallow wrappers, silent fallbacks over unclear invariants, logic in the wrong layer. Use for "code judo", "thermo-nuclear review", a deep…
Skill Claude CodeCodex
Hill-climb a product's UX and visual design to a first-party ship bar: build the design as a clickable multi-device prototype with a scenario switcher, shoot fixed evidence every round (screen sheet of every screen × state, bot-played flow recordings across devices side by side, automatic checks for tap targets /…
Skill Claude CodeCodex
Explain how a part of the codebase works before changing it: data flow, ownership, where state lives, the seams, and the gotchas, at the level a senior engineer needs to build a working mental model. Uses parallel read-only explorers for subsystems that span many files. Use for "how does X work", "walk me through"…
Skill Claude CodeCodex
Adversarial multi-model review of a change: three reviewers from different model families (Claude fable/opus plus Codex) read the same diff and rubric independently, then a lead synthesizes consensus, lone findings and disagreements into a verdict. Use for "interrogate", "tear this apart", "stress test this", risky…
Skill Claude CodeCodex
End-of-session learning pass: three reviewers read this session's transcript from different angles, a synthesizer turns their findings into concrete edits to specific skills, CLAUDE.md or scripts, and anything a script, hook or test could enforce goes to a structural backlog instead of more prose. Nothing is applied…
Skill Claude CodeCodex
Keep an append-only decision log (TSV, one row per decision: what, why, evidence pointer, result) for long, overnight or unattended runs, audit it against the transcript at the end, and have a different model family flag what the user should look at. Use for nightshift, swarm, hill-climbs, any run the user will review…
Skill Claude CodeCodex
Give a project its own scripted proof that the real app works: generate a project-local verify- skill (Launch, Doctor, Drive, Evidence, Cleanup + a feature map of every role and screen), prove it once end to end, and keep it honest with a maintenance pass. Ships av-verdict.mjs, which fails a recorded clip on…
Skill Claude CodeCodex
Summarize the user's own commits across /src repos for a time window (yesterday, this week, since Monday) into a short status update or morning report. Use for "what did I get done", "what shipped this week", "morning report", or standup-style summaries.
Skill Claude CodeCodex
Find out why code is the way it is: the commits, PRs, issues, chat threads, memory notes and incidents that shaped it, with honest confidence levels for every claim. Use for "why does X work this way", "why did we pick Y", "was this intentional", regressions ("when did this break"), before removing something that…
Skill Claude CodeCodex
Generic improvement loop for anything judged by quality rather than one number: a prompt, copy, a component, a level, a skill, an API, a scene. Freeze a rubric and an evidence command, prove the evaluators can tell bad from good, then loop: one change → fresh evidence → independent evaluators (fresh each round, mixed…
Skill Claude CodeCodex
Make a web game OGS-compatible, new or existing (Rocket Crew, Night Flight, Trivia Jam…): the catalogue entry and art kit, the TV page framed by the OGS TV launcher (ogs:start, ogs:suspend / ogs:resume, sitting labels), the phone page in the OGS app's WebView (OGS name and avatar, no name form), server-side token…
Skill Claude CodeCodex
Install an SRE agent on a service repo: a GitHub Actions cron job that reads the service's production logs every 15 minutes, groups errors by fingerprint, files or updates deduplicated GitHub issues, reopens regressions, and (when allowed) has Claude attempt a test-first fix PR. Cloudflare Workers Logs and TestFlight…
Skill Claude CodeCodex
Local routine that runs improve-codebase-architecture unattended on Jon's recently active repos: once a week per repo it implements the strongest deepening opportunity test-first as a draft PR and emails Jon the link from [email protected]; every day it revises open proposals Jon commented on. Past proposals' PR comments…
Skill Claude CodeCodex
Homegrown client telemetry for browser, React Native/Expo and Swift apps on Cloudflare: errors, session lifecycle and product events buffered on the device, posted to the project's own Worker, and re-emitted as wide events in Workers Logs, where /sre-agent files issues. Includes a tested TypeScript reference (client…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: