AgentSkillsCN

anthropic-best-practices-update

按照既定顺序（获取、归一化、拆分、提取、翻译）运行完整的 Anthropic 最佳实践流程，充分利用现有技能。当被要求一次性更新 docs/best-practices/claude-md.md 时，可直接采用此方案。

SKILL.md
--- frontmatter
name: anthropic-best-practices-update
description: Run the full Anthropic Best Practices pipeline in order (fetch, normalize, split, extract, translate) using existing skills. Use when asked to update docs/best-practices/claude-md.md in one shot.

Anthropic Best Practices Update

Overview

Execute the end-to-end Anthropic Best Practices update workflow in a deterministic order. Includes translation to Japanese via GPT-5.

Quick Start

If not already in tmux, the script will create a temporary tmux session for translation:

bash
skills/anthropic-best-practices-update/scripts/run_pipeline.sh

Skip translation:

bash
skills/anthropic-best-practices-update/scripts/run_pipeline.sh --skip-translate

Workflow

  1. •Fetch sources (doc-fetcher).
  2. •Normalize snapshots (md-normalizer).
  3. •Split into sections (md-section-splitter).
  4. •Convert normalized markdown (md-converter).
  5. •Translate to Japanese (md-translator).

Inputs

  • •data/doc-fetcher/state.json
  • •The pipeline uses a built-in list of Anthropic best-practices URLs in skills/anthropic-best-practices-update/scripts/run_pipeline.sh. Source ids are derived from URLs.

Outputs

  • •data/doc-fetcher/snapshots/
  • •data/doc-fetcher/normalized/
  • •data/doc-fetcher/sections/
  • •data/doc-fetcher/generated/claude-md.en.md
  • •docs/best-practices/<source-id>.md

Options

  • •--skip-translate: run everything except GPT-5 translation.
  • •--id <source-id>: run a single source from the built-in list (id is URL-derived).

Notes

  • •OPENAI_API_KEY が env にある場合は tmux なしで直接翻訳する。
  • •env が無い場合は tmux + 1Password (op) を使う。
  • •tmux 翻訳が止まったら、出力されたセッションに attach する。