media-generation-transcription
Skill0 starsChanged 7 months agoApache-2.0View on GitHub
Guardrail for choosing and packaging OpenAI media workflows across Sora video, speech generation, and audio/video transcription with explicit artifact discipline and blocked-state handling.
- Reads credentials
In this repository57 other files it gives its agents.
- ai-app-integrationSkill
- anti-llm-frontend-designSkill
- aspnet-core-web-developmentSkill
- async-python-patternsSkill
- browser-game-foundationsSkill
- browser-local-qaSkill
- build-mcp-server-sdk-v2Skill
- canva-design-opsSkill
- check-completionSkill
- cli-tool-creatorSkill
- cloudflare-agents-mcpSkill
- cloudflare-durable-objectsSkill
- cloudflare-workers-platformSkill
- deployment-preview-workflowsSkill
- docx-document-artifactsSkill
- durable-workflow-orchestrationSkill
- dynamic-og-imagesSkill
- eval-auditSkill
- evaluate-ragSkill
- figma-design-to-code-rulesSkill
- fine-tuningSkill
- frontend-product-builderSkill
- game-asset-pipeline-expansionSkill
- game-ui-playtestSkill
- generate-synthetic-dataSkill
- github-pr-ci-opsSkill
- gmail-inbox-opsSkill
- golden-datasetSkill
- huggingface-llm-trainerSkill
- javascript-monorepo-buildsSkill
- linear-issue-managementSkill
- millrace-rs-contract-paritySkill
- minecraft-mod-qa-automationSkill
- notion-workspace-flowsSkill
- openai-docsSkill
- optimize-agentic-cliSkill
- pdf-artifact-workflowSkill
- phaser-2d-game-developmentSkill
- postgres-query-performanceSkill
- product-ui-polish-reviewSkill
- pydantic-v2-patternsSkill
- python-repo-refactoring-hygieneSkill
- rag-corpus-hygieneSkill
- react-next-performance-reviewSkill
- rust-best-practicesSkill
- shadcn-ui-compositionSkill
- skill-evaluation-harnessSkill
- tauri-desktopSkill
- tauri-mcp-cliSkill
- tauri-ptySkill
- tile-pixel-art-transitionsSkill
- train-sentence-transformersSkill
- typer-cli-patternsSkill
- using-dbt-for-analytics-engineeringSkill
- validate-evaluatorSkill
- visual-asset-pipelineSkill
- watchdog-file-watcher-patternsSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 10 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 49
Before any live call, check OPENAI_API_KEY and access.
- Distinctive: line 55
Never ask the user to paste secrets.
- Distinctive: line 56
Treat missing key, missing access, or unavailable job as a blocked state, not a hidden failure.
- Distinctive: line 57
Keep speech augmentation short and labeled.
- Distinctive: line 59
If a request spans modes, separate outputs by name and path.
- Distinctive: line 60
Use the supported media execution path for the task instead of inventing ad hoc shell commands.
- Distinctive: line 61
Keep the response compact and decision-oriented unless the user explicitly asks for deeper guidance.
- Distinctive: line 79
Call out blocked states plainly, with the missing key, access, or job noted.
- Distinctive: line 88
If anything required for a live call is missing, stop and report the blocked state instead of guessing.
- Distinctive: line 90
Verify that the outputs match the chosen mode, the local artifacts are separated, and the blocked-state explanation is faithful.
- Distinctive: line 109
Start with the smallest useful decision: which media mode owns the request and whether access is already available.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Media Generation Transcription
Purpose
Guardrail for OpenAI media workflows across video generation, speech generation, transcription, and combined packages. Keep the mode choice explicit, the artifact tree deterministic, and blocked states honest. This is a guardrail, not a media manual.
Use When
- the task asks to generate, edit, extend, poll, or download Sora video
- the task asks to generate narration, voiceover, or other spoken audio
- the task asks to transcribe audio or video, with or without speaker labels
- the task spans more than one media mode and needs local artifact separation
- the task needs a clear blocked-state explanation for missing key, missing access, or unavailable job
Do Not Use When
- the work is generic media editing outside OpenAI media workflows
- the prompt is only creative direction with no generation or transcription action
- another more specific skill already owns the same surface
- the request would require the user to paste secrets or credentials into chat
Quick Start
- Classify the request first as
video generation,speech generation,transcription, orcombined media package. - For video, name the exact intent up front:
create,edit,extend,poll, ordownload. - For speech, keep the text verbatim and decide single clip vs batch only when multiple prompts or files are actually requested.
- For transcription, choose
text,json, ordiarized_jsondeliberately and surface speaker hints, uncertainty, and segment boundaries. - Before any live call, check
OPENAI_API_KEYand access. If either is missing, stop and report a blocked state. - For combined work, keep each output separate and preserve downloads locally before URLs expire.
- End with the next verifier and the evidence captured.
Operating Constraints
- Never ask the user to paste secrets.
- Treat missing key, missing access, or unavailable job as a blocked state, not a hidden failure.
- Keep speech augmentation short and labeled; do not rewrite the user's text.
- For Sora, enforce the guardrails: no real people, no copyrighted characters or music, and no human-face input images.
- If a request spans modes, separate outputs by name and path; do not blur a transcript, a voiceover, and a video asset into one artifact.
- Use the supported media execution path for the task instead of inventing ad hoc shell commands.
- Keep the response compact and decision-oriented unless the user explicitly asks for deeper guidance.
Inputs This Skill Expects
- the user request and any existing media asset links or file paths
- whether the work is video, speech, transcription, or combined
- exact text for speech requests
- audio or video source paths for transcription requests
- known speaker hints, output-format preferences, or naming requirements
- any stated retention or download requirements
Output Contract
- Name the media mode first.
- State the chosen branch and why it fits.
- Preserve exact text for speech requests.
- Preserve speaker hints and uncertainty explicitly for transcription.
- Separate combined outputs into named artifacts and a local manifest or report.
- Call out blocked states plainly, with the missing key, access, or job noted.
- End with the next verifier and the evidence captured.
Procedure
- Classify the media work and decide whether it is single-mode or combined.
- For video, choose
create,edit,extend,poll, ordownloadbefore writing the prompt. - For speech, preserve the input text exactly and only add short, labeled direction if the task needs it.
- For transcription, choose the response format deliberately and state what is known versus uncertain.
- If anything required for a live call is missing, stop and report the blocked state instead of guessing.
- For combined packages, create separate named outputs and a local manifest or report listing each file and its purpose.
- Verify that the outputs match the chosen mode, the local artifacts are separated, and the blocked-state explanation is faithful.
Pitfalls And Gotchas
- Collapsing video, speech, and transcription into one generic media workflow
- Asking for secrets instead of telling the user to set them locally
- Rewriting speech text instead of preserving it verbatim
- Hiding transcription uncertainty or speaker ambiguity
- Letting downloaded assets expire before they are captured locally
- Combining unrelated outputs without a manifest or report
- Using a broad media tutorial tone instead of a guardrail tone
Rejected Trope
- Rejected trope: split video, speech, and transcription into separate skills in v1.
- Better alternative: keep one umbrella guardrail because the mode selection, credential preflight, artifact handling, and blocked-state rules are shared.
Progressive Disclosure
Start with the smallest useful decision: which media mode owns the request and whether access is already available. Expand only enough to name the branch, the artifacts, and the next verifier. Keep the body short enough that it steers decisions instead of becoming a manual.
Verification Pattern
- Confirm the mode choice is explicit.
- Confirm access preflight happened before any live call.
- Confirm speech text stayed verbatim and transcription format was chosen deliberately.
- Confirm combined jobs have separate artifact names and a local manifest or report.
- Confirm blocked states are clearly stated when key or access is missing.
- Next verifier: checker or execution reviewer; capture the chosen branch, the access status, and the artifact tree or download set.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

