brandonburrus Purpose
Consolidated entry point for language and framework best practices. Detect the stack in scope, then load only the reference files the task needs; the practices live in references/ precisely so that unrelated stacks never consume context. Loading every reference defeats the design.
Project discovery (always do first)
- Detect frameworks and tooling from manifests and configs:
package.jsondependencies,vitest.config.*,playwright.config.*,cdk.json,*.tffiles,go.mod,Cargo.toml,pyproject.toml. - Detect the language from the extensions of the files actually being touched.
- Load references per the routing table below.
Routing table
| Context | Load |
|---|---|
| React component or app work | React + TypeScript or JavaScript by file type |
| Node.js service, CLI, or module work | Node + TypeScript or JavaScript |
| Vitest unit or integration tests | Vitest + TypeScript or JavaScript |
| Playwright E2E tests | Playwright + TypeScript or JavaScript |
| AWS CDK infrastructure | CDK + TypeScript |
| Terraform infrastructure | Terraform |
| TypeScript, no framework match | TypeScript |
| JavaScript, no framework match | JavaScript |
| Python | Python |
| Go | Go |
| Rust | Rust |
| SQL queries or migrations | SQL |
| Bash or shell scripts | Bash |
| pandas data work | pandas + Python |
| NumPy numerical code | NumPy + Python |
| polars data work | polars + Python |
| marimo notebooks | marimo + Python |
| Resilience and fault-tolerance work (retries, timeouts, idempotency, circuit breakers, fallbacks, queue consumers, health checks) | Resilience + the language reference for the files touched |
| Jupyter notebooks | Jupyter + Python |
Loading rules
- When a framework applies, load its framework reference first, then exactly one language reference for the files being edited.
- For mixed-language changes, load one language reference per touched language.
- Never load references "just in case"; an unloaded reference costs nothing, an unused loaded one crowds out the task.
Precedence and boundaries
- On conflict, the project's own documented conventions win, then the global CLAUDE.md rules, then these references. A reference here never justifies diff noise in a codebase that does it differently.
- The SQL reference covers writing queries and migrations; designing tables, keys, and indexes belongs to the spec skill (schema reference).
- Endpoint and schema contracts belong to the spec skill (api reference); these references inform the implementation behind the contract.
- This skill informs how code is written, not the workflow around it: test-first discipline is follow-tdd, behavior-preserving cleanup is refactor, debugging is fix.