K8s Self Hosted Whisper Api

Transcribe audio via the self-hosted Whisper ASR instance running on Kubernetes. Use this skill whenever the user wants to transcribe audio files, convert sp...

Security Scan

The skill is internally consistent with its stated purpose: an instruction-only helper that posts an audio file to a self-hosted Whisper ASR service on a Kubernetes DNS name and writes the returned transcript; nothing in the package appears disproportionate or unrelated to transcription.

Purpose & Capability

Name, description, SKILL.md, and the included transcribe.sh all align: they send an audio file to the cluster-local Whisper ASR /asr endpoint and save the transcript. Required binary (curl) matches the script's primary network calls.

Instruction Scope

The instructions and script only read the specified audio file and write the transcript output. They POST the file to the declared cluster-local endpoint and do not reference unrelated system files or external endpoints. Two notes: (1) the SKILL.md and script state 'No authentication required' — this means sensitive audio will be sent to that service without credentials, so users should confirm they trust that service. (2) the script invokes python3 to URL-encode an initial prompt (fallback is provided), but python3 is not listed in required binaries.

Install Mechanism

This is an instruction-only skill with no install spec and only a small shell script; nothing is downloaded or written to disk by an installer—low install risk.

Credentials

No environment variables, credentials, or config paths are requested. The script uses a hard-coded cluster-local base URL; no extra secrets are required, which is proportionate for the stated purpose.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system-wide configs. It is user-invocable and can be invoked autonomously (platform default), which is expected for a skill of this type.

Assessment

This skill is coherent for transcribing audio to a self-hosted Whisper service, but before installing or using it you should: (1) confirm the cluster-local endpoint (http://whisper-asr.whisper-asr.svc.cluster.local:9000) is the intended and trusted service — audio data will be uploaded there without authentication; (2) ensure the agent environment can reach that DNS name (it's a Kubernetes in-cluster name and may be unreachable from outside the cluster); (3) verify python3 is available if you plan to use the --prompt flag (the script falls back but may behave differently); and (4) treat transcripts/audio as potentially sensitive and ensure they are handled per your privacy policy. If any of these are unacceptable, do not enable the skill.

Self-Hosted Whisper API (curl)

Transcribe an audio file via the Whisper ASR webservice at http://whisper-asr.whisper-asr.svc.cluster.local:9000.

Uses the onerahmet/openai-whisper-asr-webservice API (/asr endpoint).

Quick start

{baseDir}/scripts/transcribe.sh /path/to/audio.m4a

Defaults:

• Endpoint: http://whisper-asr.whisper-asr.svc.cluster.local:9000/asr
• Task: transcribe
• Output: txt

Useful flags

{baseDir}/scripts/transcribe.sh /path/to/audio.ogg --language en --out /tmp/transcript.txt
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --language de
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --json --out /tmp/transcript.json
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --output srt --out /tmp/subtitles.srt
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --output vtt
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --translate
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --vad-filter --json
{baseDir}/scripts/transcribe.sh /path/to/audio.m4a --word-timestamps --json

Notes

• Supported --output formats: txt, json, vtt, srt, tsv
• --translate produces an English transcript regardless of source language
• --vad-filter enables voice activity detection to skip silent sections
• --word-timestamps adds word-level timing (use with --json)
• The model is configured on the server side (ASR_MODEL env var), not per request
• Swagger docs available at http://whisper-asr.whisper-asr.svc.cluster.local:9000/docs
• No authentication required