Opinionated conventions for authoring, restructuring, and reviewing agent skills. Use when deciding whether guidance belongs in `SKILL.md`, a conditional reference, or a smaller skill; writing trigger descriptions and reference routers; naming procedural or documentary skills; creating examples; keeping skill context focused; or bundling static assets and their licenses.
Opinionated conventions for designing focused agent-skill plugins as encapsulated, installable capabilities. Use when defining a plugin boundary, deciding which skills are public entry points, hiding shared implementation skills, splitting broad plugins, designing plugin structure, minimizing client manifests, or removing auxiliary package documentation.
Opinionated language-agnostic design rules for implementation, refactoring, and code review. Use when shaping conditional control flow, finite states, invariants, error boundaries, resource lifetimes, abstractions, public APIs, validation boundaries, pagination, missing values, or the separation of sans-I/O decisions from imperative drivers.
Opinionated language-agnostic test-admission and sans-I/O unit-testing rules. Use when proposing, writing, reviewing, or deleting tests; deciding whether behavior is project-owned; evaluating wrapper, schema, ORM, framework, mock, or third-party-library tests; or separating pure business policy from wiring and I/O.
Opinionated feature-first architecture conventions for cohesive modules, owned subtrees, entry points, shared-code promotion, and package boundaries. Use when designing or reviewing directory structure, placing UI or workflow code, splitting nested capabilities, extracting shared modules, enforcing feature import boundaries, deciding whether a feature deserves a package, or refactoring horizontal technical layers.
Opinionated `better-all` conventions for dependency-declared async task graphs, inferred task results, cancellation propagation, and failure semantics. Use when replacing serial `await` chains or manual `Promise.all` stages, parallelizing work with result dependencies, or writing and reviewing `all` and `allSettled` task definitions.
Opinionated logging and error-handling conventions for OpenTelemetry-compatible systems covering log levels, attribute naming, and error classification. Use when writing log statements, choosing log levels, or handling errors with traces.
Opinionated Python 3.13+ conventions for application and library code. Use when writing or reviewing Python typing, protocols, generics, finite states, optional values, exceptions, serialized-data validation, package exports, import layout, typed distributions, async resource ownership, async pagination, or per-distribution packaging metadata.