Skip to content

README.md

Latest commit

 

History

History
21 lines (13 loc) · 1.82 KB

README.md

File metadata and controls

21 lines (13 loc) · 1.82 KB

claude-dev-toolkit

Set up Claude Code well and build with it.

  • starter-kit/ — drop-in CLAUDE.md + MEMORY.md + ERRORS.md + anti-style.md, plus a .claude/ config tree for autonomous work (permission allowlist, /loop + /goal commands, an orchestrator subagent), with an install.py (skips existing files) and tests. The highest-leverage thing for Claude Code quality.
  • long-running-autonomy.md — de-hyped writeup of the five tips for running Opus 4.8 autonomously on long jobs (auto-permissions, subagent orchestration, /loop+/goal, cloud, self-verify), mapped to the files that bake them into every project.
  • secrets-manager/ — keep API keys out of Claude Code's on-disk JSONL transcripts: a setup guide + a stdlib secret_env.py helper (with tests).
  • frontend-polish/ — a screenshot → self-review → refine loop that composes the official frontend-design skill instead of reinventing aesthetics.
  • a2a-agent-tutorial/ — a minimal, dependency-light Agent2Agent (A2A) demo in Python (pluggable LLM, runs offline, tested).
  • commands-reference.md — which "Claude commands" are real native features vs. just renamed prompts.

Run any project's tests with python -m pytest <subfolder>.

Use in Claude chat (claude.ai)

These are Claude Code / runtime tools (installers, a secrets helper, a Playwright screenshot loop, runnable agent code) — they need a local environment, so they aren't claude.ai chat skills. The chat-usable skills live in the sibling repos claude-thinking-tools, claude-writing-tools, and claude-everyday-tools (see their skills/).

Curated/built from techniques shared by creators on Instagram (May 2026); marketing hype and inflated stats stripped out. Prompt text is rewritten, not copied; source handles are credited in the pack files.