Skill Claude CodeCodex
Stamp every serialized artifact with a schema version and evolve the format over time without breaking data that was already persisted — detect the version on load, migrate old shapes forward, and keep readers for old versions working. Use this whenever you design or change a serialization format (JSON, SQLite rows…
Skill Claude CodeCodex
Treat "zero trust" as a per-request property to be verified, not a product you bought or a network you segmented — the honest question is always "trusted to do what, verified how, on every request", and the failure is the implicit trust that survives the architecture diagram: a flat network behind the VPN, a service…
Skill Claude CodeCodex
Derive cloud infrastructure from a workload's actual constraints and required capabilities — never from "X is popular" or from whichever provider is already open in a tab. Use whenever choosing or re-evaluating a cloud provider, a managed service, or a deployment architecture; whenever comparing…
Skill Claude CodeCodex
Priority determines resolution order, not permission to accumulate indefinitely — every known issue (bug, inconsistency, TODO, workaround, deferred P3) must terminate in resolved, explicitly rejected with a reason, or documented as an open unresolved constraint, never silently vanish by being written down. Use…
Skill Claude CodeCodex
Build toward a chosen, ambitious final product in coherent levels — each one architecturally compatible with the destination, useful standalone, and carrying the invariants (security, persistence, authority boundaries) it will need later, never bolted on afterward. Use whenever planning or building a project meant to…
Skill Claude CodeCodex
Design the intermediate state, redelivery, and deduplication explicitly whenever two effects cannot share one transaction — a DB write plus an event/webhook/second-service call, a payment plus a notification, a state change plus a cache invalidation. Use whenever an operation must affect two systems that don't share a…
Skill Claude CodeCodex
Build interactive frontend components — forms, modals, navigation, custom widgets — so they are operable and perceivable by construction, not patched afterward with ARIA. Use whenever writing or reviewing a component with keyboard interaction, focus behavior, or non-text content. Trigger on "accessible", "a11y"…
Skill Claude CodeCodex
Decide where to cut a frontend component's boundary by cohesion, behavior, semantics, reuse, and shared reasons to change — not by a mechanical "one responsibility per component" rule that produces ceremonial fragmentation. Use whenever creating a new component, splitting or merging existing ones, or reviewing a PR…
Skill Claude CodeCodex
Set and enforce explicit performance budgets before writing frontend code, not measure after the fact — outcome metrics (LCP/INP/CLS) and the causal resource budgets that actually produce them (JS shipped, image weight, font loading, third-party scripts, critical request count, hydration cost). Use whenever adding a…
Skill Claude CodeCodex
Allocate frontend test effort by the Testing Trophy (static > unit > integration > E2E) and write tests that assert on what the user perceives, using stable selectors, instead of internal implementation. Use whenever writing or reviewing UI tests, deciding what layer to test something at, or diagnosing flaky tests.…
Skill Claude CodeCodex
Change a request/response shape, RPC contract, or message schema so that vN and vN+1 producers/consumers keep working while both are simultaneously alive — not just so a future reader can parse an old artifact. Use whenever changing a field, endpoint, protobuf message, or GraphQL type during or across a rolling…
Skill Claude CodeCodex
Design and review user flows, forms, dashboards, and onboarding for cognitive, sensory, and choice load — grounded in WCAG 2.2 cognitive-accessibility criteria and W3C COGA guidance, and applicable to every user, not only neurodivergent ones, because reduced cognitive load helps anyone under stress, time pressure, or…
Skill Claude CodeCodex
Design every view that fetches or mutates data around its full state machine — loading, success, empty, and error are all mandatory, none optional — with error handling that fails at the right boundary and produces a specific, actionable message. Use whenever building or reviewing a component that does async data…
Skill Claude CodeCodex
Write an AGENTS.md (or CLAUDE.md) that actually changes agent behavior — falsifiable, enforced invariants grounded in this repo's real incidents, not a generic checklist any repo could have. Use whenever creating or reviewing a repo's agent-facing contract file, whenever an agent keeps making the same mistake a doc…
Skill Claude CodeCodex
Present a technical project out loud, under a real time limit, to an audience that may not share your vocabulary or technical background — without hyping unearned claims, without opening on caveats, and without inventing an answer live rather than saying "I don't know." Use whenever preparing or delivering a pitch…
Skill Claude CodeCodex
Severity controls response urgency; it does not authorize neglect. A P0/P1 earns immediate review — but P2/P3/P4 remain known defects requiring disposition, not a silent backlog, and before anything is deferred its severity must be re-tested under interaction, composition, adversarial control, and deployment context…
Skill Claude CodeCodex
Select programming languages, DSLs, query languages, policy languages, proof systems, or runtimes from the concrete forces of the problem rather than from model familiarity or the human's prior exposure. Use whenever new code must be written and the language is not already fixed: scripts, tools, services, parsers…
Skill Claude CodeCodex
Order a complex technical project's presentation (a README, a hackathon submission, a project landing doc) by the reader's cognitive dependency, not by architectural importance — problem before mechanism, observable behavior before internals, evidence before technical depth. Use whenever writing or reviewing a README…
Skill Claude CodeCodex
Minimize the user's executive-function cost when operating computers, services, dashboards, forms, CLIs, or administrative workflows on their behalf — the user is technically fluent (terminal, Linux, SSH, APIs, scripts) but has high assistance needs (ADHD/low-spoon days, or anyone under cognitive load) and needs every…
Skill Claude CodeCodex
Default to working directly on main on the user's own solo repositories — including with multiple agents active on main at once, kitchen-line style, each owning its piece and none assuming another's work is correct — instead of a feature-branch-per-agent/per-change workflow that is actually harder to track with ADHD…
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: