菜单
Transcribe local audio with Gemini, including speaker labels and timestamps. Use for podcasts, interviews, lectures, meetings, voice notes, or requests to convert MP3, WAV, M4A, OGG, FLAC, or WebM audio into Markdown.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Restate the previous response plainly and concisely. Use only when the user explicitly asks for a simpler human explanation or invokes bro.
Apply correct-by-construction TypeScript engineering standards covering boundary parsing, typed failures, domain modeling, module boundaries, dependency injection, testing, and type safety. Use when writing or reviewing TypeScript.
Audit installed Agent Skills for duplicate copies, unused candidates, oversized descriptions, loaded roots, and prompt-budget pressure. Use when cleaning or debugging skill collections across common agent runtimes.
Apply concrete clarity, concision, structure, grammar, and plain-English checks when editing prose, reviewing drafts, writing reports or articles, or teaching clear writing.
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
Produce an implementation-ready technical specification using typed contracts, concrete module boundaries, call stacks, data flows, alternatives, file changes, and vertical red-green test slices. Use when the user asks for a tech spec or architecture handoff.
| name | audio-transcribe |
| description | Transcribe local audio with Gemini, including speaker labels and timestamps. Use for podcasts, interviews, lectures, meetings, voice notes, or requests to convert MP3, WAV, M4A, OGG, FLAC, or WebM audio into Markdown. |
| license | MIT |
| compatibility | Requires Bun 1.1+, ffmpeg/ffprobe, network access, and a Gemini API key. |
The bundled CLI normalizes audio, splits long recordings, transcribes chunks with Gemini, and writes one Markdown transcript.
Resolve paths relative to this SKILL.md; do not assume a global skill directory.
bun install --cwd <skill-directory>/scripts --no-save
export GEMINI_API_KEY='...'
GOOGLE_API_KEY and OPENCODE_GOOGLE_API_KEY are accepted as fallbacks. Set GEMINI_TRANSCRIBE_MODEL to override the default model.
ffprobe to inspect duration.Do not upload media without the user's explicit transcription request.
bun run <skill-directory>/scripts/transcribe.ts \
./interview.m4a \
--output ./interview-transcript.md
Options:
--chunk-minutes N: chunk duration; default 20--output PATH: exact Markdown output path--keep: retain normalized audio and chunk files beside the transcript--help: show usageSupported inputs: MP3, WAV, M4A, OGG, FLAC, and WebM.
The transcript contains source metadata, duration, model, chunk count, speaker-labelled turns, and timestamps. Chunk prompts receive their offset within the full recording so timestamps remain useful across long files.
Speaker identity inferred by a model can be wrong. Preserve generic labels when names are uncertain and do not present inferred identities as verified facts.
After transcription:
The CLI does not write partial transcripts after an API failure. Temporary normalized audio is removed unless --keep is set.
Never print API keys or add private media and generated transcripts to the skill repository.