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 calesthio/generative-media-skills --skill openai-audiogit clone --depth 1 https://github.com/calesthio/generative-media-skillsWrote 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/calesthio/generative-media-skills/openai-audio)<a href="https://agentmods.dev/skills/calesthio/generative-media-skills/openai-audio"><img src="https://agentmods.dev/badge/skills/calesthio/generative-media-skills/openai-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/calesthio/generative-media-skills/openai-audio"><img src="https://agentmods.dev/badge/skills/calesthio/generative-media-skills/openai-audio.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 4 findings, up to medium
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 →
- medium Output Handling · line 19 Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.Fix: Set explicit limits on output length, generation count, and rate. Use max_tokens and truncation to prevent unbounded output.
- medium Excessive Agency · line 27 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Excessive Agency · line 113 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Excessive Agency · line 458 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00081 | $0.07403 |
| Opus 5 | $0.00041 | $0.03701 |
| Sonnet 5 | $0.00016 | $0.01481 |
| Haiku 4.5 | $0.00008 | $0.00740 |
Grade A, and why
openai-audio scanned grade A with 0 findings 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 9d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 591 lines — stays where its author put it; the contents beside it link to each section on GitHub.
OpenAI audio
Use this skill when a user wants OpenAI audio production or understanding in a bounded request: generated narration, spoken product demos, podcast or interview transcription, subtitle timing, translation to English, audio QA, or adding audio input/output to an existing chat workflow.
Do not use this as the primary guide for continuous low-latency speech-to-speech agents, live interpreting, SIP/WebRTC sessions, or voice activity detection. Those belong to the separate realtime voice skill. This skill may still explain the boundary: request-based audio APIs fit files, scripts, and bounded requests; realtime sessions fit open connections with live audio events. Documented source: OpenAI’s audio guide distinguishes request-based APIs, realtime sessions, and multimodal chat completions; verified 2026-07-10: https://developers.openai.com/api/docs/guides/audio
Source status and evidence labels
Treat the following as documented facts verified on 2026-07-10 from OpenAI public docs:
- The request-based speech endpoint is
POST /v1/audio/speech; it acceptsgpt-4o-mini-tts, the current pinned snapshotgpt-4o-mini-tts-2025-12-15,tts-1, andtts-1-hd, and supports built-in voices plus custom voice IDs where available. Speech input is limited to 4,096 characters. The older pinned snapshotgpt-4o-mini-tts-2025-03-20is scheduled to shut down on 2026-07-23 withgpt-4o-mini-tts-2025-12-15as its documented substitute; do not confuse that snapshot retirement with retirement of the unpinned alias. Sources: https://developers.openai.com/api/reference/resources/audio/subresources/speech/methods/create and https://developers.openai.com/api/docs/deprecations gpt-4o-mini-ttssupports instruction-style voice control; the oldertts-1andtts-1-hdmodels do not support theinstructionsparameter. Source: https://developers.openai.com/api/reference/resources/audio/subresources/speech/methods/create- Speech output formats for
/v1/audio/speechare documented asmp3,opus,aac,flac,wav, andpcm;speedranges from0.25to4.0; streaming format can besseoraudio, with SSE not supported fortts-1/tts-1-hd. Source: https://developers.openai.com/api/reference/resources/audio/subresources/speech/methods/create - The speech-to-text endpoint is
POST /v1/audio/transcriptions; documented models includegpt-4o-transcribe,gpt-4o-mini-transcribe,gpt-4o-transcribe-diarize, andwhisper-1, with response-format and streaming differences by model. File uploads are limited to 25 MB; split or compress larger inputs without cutting mid-sentence. Source: https://developers.openai.com/api/docs/guides/speech-to-text gpt-4o-transcribe-diarizedoes not supportprompt, logprobs, ortimestamp_granularities[]; input longer than 30 seconds requireschunking_strategy, with"auto"recommended. Source: https://developers.openai.com/api/docs/guides/speech-to-text- Transcription streaming is supported for current GPT transcription models, but not for
whisper-1; streamed transcription emits transcript delta/done events. Source: https://developers.openai.com/api/docs/guides/speech-to-text - The translation endpoint is
POST /v1/audio/translations; the API reference states that onlywhisper-1is currently available for translation to English, even though shared model type enums may list additional audio model IDs. Source: https://developers.openai.com/api/reference/resources/audio/subresources/translations/methods/create - Audio-capable chat completions can use
gpt-audio-1.5withmodalities: ["text", "audio"], anaudioobject for voice/format, andinput_audiocontent blocks for audio input. Source: https://developers.openai.com/api/docs/guides/audio - Realtime routing is separate: the realtime overview points low-latency voice agents to
gpt-realtime-2.1, live translation togpt-realtime-translate, and live transcription togpt-realtime-whisper. Source: https://developers.openai.com/api/docs/guides/realtime - Pricing is volatile. As of 2026-07-10, OpenAI’s pricing page lists realtime/audio-generation prices by token or minute and transcription prices for
gpt-4o-transcribeandgpt-4o-mini-transcribeby token with estimated per-minute costs. Always re-check before quoting costs in production. Source: https://developers.openai.com/api/docs/pricing - OpenAI’s platform data-controls page says API endpoint data is not used for training, lists abuse-monitoring and application-state retention by endpoint, and says
/v1/audio/transcriptionsand/v1/audio/translationshave no abuse-monitoring retention while/v1/audio/speechhas 30-day abuse-monitoring retention by default; verified 2026-07-10. Source: https://developers.openai.com/api/docs/guides/your-data - OpenAI’s usage policies prohibit use of someone’s likeness, including voice, without consent in ways that could confuse authenticity. The TTS guide also requires clear disclosure to end users that the voice is AI-generated, and custom-voice creation is limited to eligible customers and requires separate matching consent and sample recordings. Sources: https://openai.com/policies/usage-policies/ and https://developers.openai.com/api/docs/guides/text-to-speech
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 9d ago First seen · 591 lines · 81 tokens per session scan A e7b2f5953d83
openai-audio is a skill published in the GitHub repository calesthio/generative-media-skills (170 stars, last pushed 2mo ago), licensed MIT. It adds 81 tokens to every session and 7,403 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
cliptalk-cover-director
Produces evidence-backed cover candidates and reviewable cover variants for a ClipTalk video. Use when the user asks for a cover, poster frame, thumbnail, or multiple cover directions; do not use for timeline editing or social-video reframing.
cliptalk-smart-reframe
Creates a subject-aware, time-varying crop track and a review-only social-format preview from an accepted ClipTalk cut. Use for automatic vertical, square, or portrait reframing; do not use for a fixed manual crop or before content editing is accepted.
cliptalk-content-extractor
Locates and assembles source passages matching a semantic request. Use for extracting explanations, topics, quotes, demonstrations, or other specifically described content.
cliptalk-interview-editor
Produces a coherent interview edit by combining speaker discovery, topic selection, dialogue context, cleanup, subtitles, and preview. Use for interviews, podcasts, testimonials, or question-and-answer recordings.
cliptalk-shortform-hook-director
Finds and assembles a reviewable short-form cut with a strong opening hook. Use for Shorts, Reels, social clips, talking-head cutdowns, or requests for a punchier opening.
cliptalk-social-reframe-exporter
Creates a review-only 9:16, 4:5, 1:1, or 16:9 version from an existing accepted ClipTalk cut, then checks the rendered preview. Use only when a cut already exists and the user asks to adapt it for Shorts, Reels, Douyin, Xiaohongshu, WeChat Channels, or square feeds; do not use when the user still needs content found…