Skip to content

CLAUDE.md

Latest commit

 

History

History
50 lines (38 loc) · 2.22 KB

CLAUDE.md

File metadata and controls

50 lines (38 loc) · 2.22 KB
Raw

Audio Transcriber — FunASR

This project is a Claude Code plugin providing Chinese audio transcription using FunASR with speaker diarization and LLM cleanup.

Project Structure

  • plugins/audio-transcriber/ — The plugin source code
  • .agents/skills/ — Autonomous dev team skills (installed via skills.sh)
  • .claude/skills/ — Symlinks for Claude Code skill discovery
  • .claude-plugin/marketplace.json — skills.sh marketplace registration
  • output/ — Generated transcription outputs (gitignored)

Development Workflow

  • Use /autonomous-dev skill for code changes (TDD, worktree isolation, CI verification)
  • Use /autonomous-review skill for PR code review before merging
  • Use /create-issue skill for structured GitHub issue creation
  • Use document-skills:skill-creator skill to verify/review SKILL.md changes
  • All skill source lives under plugins/audio-transcriber/skills/audio-transcribe/
  • Scripts are in scripts/, references in references/
  • The main entry point is SKILL.md
  • Run tests: cd plugins/audio-transcriber/skills/audio-transcribe/scripts && python3 -m pytest test_speaker_verification.py -v

Conventions

  • English for code, comments, commit messages, and documentation
  • Chinese audio content is the primary use case but the pipeline is language-agnostic
  • Keep SKILL.md lean (<500 lines); move details to references/
  • Scripts run directly from plugin directory via $SCRIPTS — never copy to CWD

Placeholder Names in Tests, Comments, and Docs

Never use real person names (podcast hosts, guests, colleagues, public figures) in test fixtures, docstrings, inline comments, or example strings. Always use generic placeholder names instead.

  • Chinese: 张三, 李四, 王五, 赵六, 张飞, 关羽, 刘备, 李雷, 韩梅梅
  • English: Alice, Bob, Carol, Dave, Eve, Host, Guest
  • Roles: use generic labels like Speaker 1, Host, Guest when a name is not needed

This applies to:

  • Test data (test_*.py)
  • Docstring examples (e.g., "主播：张三")
  • Inline comments illustrating input/output
  • Sample files under references/ or docs/

When a real name appears in a user-provided show notes or audio file during development, strip it before checking anything into the repo.