Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add montezuma-p/harken --skill transcribe-audiogit clone --depth 1 https://github.com/montezuma-p/harkenWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/montezuma-p/harken/transcribe-audio)<a href="https://agentmods.dev/skills/montezuma-p/harken/transcribe-audio"><img src="https://agentmods.dev/badge/skills/montezuma-p/harken/transcribe-audio/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/montezuma-p/harken/transcribe-audio"><img src="https://agentmods.dev/badge/skills/montezuma-p/harken/transcribe-audio.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 2 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Tool Misuse · line 18 Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.Fix: Limit tool chaining depth and validate the output of each tool before passing it to the next. Require explicit user approval for multi-step chains.
- low Supply Chain · line 17 Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.Fix: Avoid downloading and executing remote scripts. Use trusted packages from PyPI/npm. If remote fetch is required, verify checksums and use HTTPS.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00063 | $0.00731 |
| Opus 5 | $0.00032 | $0.00365 |
| Sonnet 5 | $0.00013 | $0.00146 |
| Haiku 4.5 | $0.00006 | $0.00073 |
Grade A, and why
transcribe-audio scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl --proto '=https' --tlsv1.2 -LsSf \ What it actually says
Transcribe Audio (harken, local-only)
Transcribe audio with harken, a local whisper.cpp CLI (single static binary). Audio never leaves the machine — do not send audio to cloud transcription APIs, and do not write inline whisper one-liners (they reload the model per file; harken loads it once per batch).
Invocation: use harken ... if the command is on PATH. If it is missing,
install it with the one-liner (no Python, no runtime deps):
curl --proto '=https' --tlsv1.2 -LsSf \
https://github.com/montezuma-p/harken/releases/latest/download/harken-installer.sh | sh
# Batch: files, dirs, or globs (model loads once per run)
harken ~/Downloads/audios/*.opus --out /path/to/transcripts
# WhatsApp export zip: no manual unzip needed; dates inclusive
# (quote paths with spaces using $HOME — a quoted ~ does not expand)
harken whatsapp "$HOME/Downloads/WhatsApp Chat - X.zip" \
--from 2026-07-13 --to 2026-07-14 --out /path/to/out --merge
- Outputs: per-file
.txt+manifest.jsonl(source, text, duration) in--out. Existing outputs are skipped;--forceredoes them. --mergewrites_chat.transcribed.txt— the chat with each voice note's transcript inlined (>> [transcript] ...).- Defaults:
--lang pt,--model small, CPU (safe everywhere, no GPU required). --lang autofor non-Portuguese audio;--model mediumwhen accuracy matters more than speed;--model small-q5_1for a 60% smaller download;--format json|srt|mdfor timestamps (mdis the readable one: one[hh:mm:ss] lineper segment).- First ever run downloads the ggml model (~466 MB for
small) to~/.cache/harken/models; after that it works fully offline.harken warm(orharken warm --model medium) pre-downloads without transcribing — useful right after installing. - Never loop over files one
harkeninvocation each: pass them all in one invocation (the model loads once per run, and per-file runs reload it). - If the harken MCP server is connected (
claude mcp add harken -- harken mcp), prefer its tools —transcribe_fileandtranscribe_whatsapp_export— over the shell invocations above; the CLI remains the fallback. Itstranscribe_statustool reports the model cache state without transcribing anything — whether the first call would download, or is already waiting on a startup download in flight. - Full docs: README.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 54 lines · 63 tokens per session scan A d846c1624463
transcribe-audio is a skill published in the GitHub repository montezuma-p/harken (13 stars, last pushed 9d ago), licensed Apache-2.0. It adds 63 tokens to every session and 731 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
transcribe-tool
Audio transcription tool. Converts audio files to text with Whisper and optional LLM post-processing. Use when: transcribing meetings, podcasts, or extracting text from recorded audio files.
audio-transcriber-transcription
Speech-to-text on the audio-transcriber MCP server — run Whisper (faster-whisper, falling back to openai-whisper) over a local audio/video file or a microphone recording, and export txt/srt/vtt/json captions. Use when the agent must transcribe or translate spoken audio, generate subtitle/caption files, or pick a…
whisper
OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual…
whisper
OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual…
video-watch
Transcribe YouTube videos and podcast episodes in high quality (Whisper via a configurable API endpoint) and distill compact findings with timestamps. Use whenever the user wants a video "watched", summarized, transcribed or analyzed, drops a YouTube or podcast URL together with a question, asks "what does he say…
dialogue-transcriber
Transcribe audio or video of conversations — interviews, panel discussions, meetings, podcasts, YouTube videos — and identify who said what (speaker diarization) using the transcriber CLI. Use this skill whenever the user wants a transcript of multi-speaker audio, asks "who said what", wants speakers labeled or…