Skip to content

references.md

Latest commit

 

History

History
91 lines (75 loc) · 5.19 KB

references.md

File metadata and controls

91 lines (75 loc) · 5.19 KB
Raw

References

External resources worth keeping within reach — agent skills, package hygiene, and language reference. Not snippets to copy; pointers to follow.

Claude Code best practices

Background reading for ai-collaboration.md. These are principles/philosophy and happy-path guidance — none is a friction catalog, which is the niche that doc fills.

  • Claude Code best practices (Anthropic, official) — principle-based, organized around context-window management; covers CLAUDE.md, /init, permission allowlisting, and worktree parallelism. States the happy path rather than the edge cases.
  • How Anthropic teams use Claude Code (Anthropic, PDF) — the most tips-flavored official artifact: team-by-team practices like detailed CLAUDE.md files and preferring MCP servers over CLIs for sensitive data. Coarser-grained than a per-gotcha log.
  • awattar/claude-code-best-practices — community aggregator of patterns with links out to practitioner write-ups (worktree parallelism, memory management, hooks, subagents).
  • rosmur claudecode-best-practices — a synthesized guide distilling ~12 sources; leans on planning-before-implementation and the observation that simple control loops tend to beat multi-agent setups.

Agentic delivery (Bryan Finster)

Bryan Finster is a well-known continuous-delivery engineer (Minimum Viable CD / minimumcd.org); these are his current, actively-maintained AI-agent delivery materials — strong on holding agent-generated code to the same quality bar as human-written code.

  • bdfinst/agentic-dev-team (docs) — three Claude Code plugins: a persona-driven dev-team (Orchestrator + specialist agents + a TDD-gated /specs → /plan → /build → /pr workflow with a reviewer swarm), plus security-assessment and marketplace-dev. A mature, real-world example of a rules / skills / agents system.
  • bdfinst/ai-patterns — a living playbook of patterns for delivering software with AI agents: agentic code review, Agentic Continuous Delivery (ACD), a defect-detection cheat sheet, contract-testing strategies, and a curated index of other skill libraries.

Rules & skills collections

Curated indexes of .claude/rules, CLAUDE.md, skills, and subagents worth mining for patterns — the community layer above the config/claude-code/rules snippets here.

  • hesreallyhim/awesome-claude-code — the de-facto meta-index: hand-picked skills, subagents, hooks, CLAUDE.md guides, status lines, and tooling. Actively maintained; the best single entry point.
  • josix/awesome-claude-md — curated exemplary CLAUDE.md files mined from real repos, with analyses and templates. Narrowly on-target when you're refining a CLAUDE.md.

Agent skills

  • anthropics/skills — the official Agent Skills repo: production document skills, a skill-creator, and the canonical SKILL.md specification (spec/agent-skills-spec.md). The authoritative source for the format.
  • Intopia web accessibility skill — agent skill giving AI coding tools WCAG-aligned accessibility guidance, code examples, and review tooling for building and auditing web interfaces.
  • Vercel agent skills — a collection of skills for AI coding agents (packaged instructions + scripts in the Agent Skills format) that extend agent capabilities.
  • Community-Access/accessibility-agents — specialist agents that enforce WCAG 2.2 AA in Claude Code (ARIA patterns, focus management, contrast, live regions, forms); honest that automated checks still need real assistive-tech verification.
  • fecarrico/A11Y.md — an a11y rules file: an 8-rule WCAG 2.2 AA behavioral contract to inject into CLAUDE.md, with tiered A/AA/AAA profiles and an on-demand APG reference. The rules-file complement to the Intopia / Community-Access skills.

Package hygiene

  • replacements.fyi — search an npm package name to find better/safer modern alternatives to deprecated or abandoned dependencies.

TypeScript

  • TypeScript tips everyone should know — curated practical TypeScript patterns (e.g. prefer unknown over any, discriminated unions, template literal types, the satisfies operator) for safer, more readable code.
    • TODO: create a Claude Code agent skill around these tips — package them as a reviewable skill that surfaces the relevant pattern during TS work.