---

name: funclip description: "Guide FunClip ASR video/audio clipping, LLM-assisted clipping providers, and release packaging workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

FunClip repo skill

Use this skill when the task names FunClip or asks for ASR-driven video/audio clipping, transcript timestamp clipping, speaker clipping, subtitle generation, LLM-assisted highlight selection, or FunClip release packaging.

FunClip is a checkout-oriented Python application. Prefer the documented script entry points (python funclip/launch.py and python funclip/videoclipper.py) instead of assuming a packaged console entry point exists.

Start here

1. Read references/installation-and-runtime.md before installing, launching, or debugging optional media/model/provider dependencies.
2. Run scripts/check_environment.py with --repo-root <funclip-checkout> when you need a no-network import, requirement, model-selection, and launch-policy smoke check.
3. Read references/troubleshooting.md for cross-cutting install/import, model-download, optional-provider, media-binary, and repo-maintenance routing failures.
4. Read references/repo-provenance.md before deciding whether this skill is stale for a checkout.

Sub-skill routes

User task	Go to	Why
Launch the Gradio app, choose Paraformer/Fun-ASR-Nano/SenseVoice, bind a remote server, run CLI stage 1/2, clip by transcript text or speaker, generate SRT/subtitle overlays, or debug no-match behavior	clip-workflows	Owns the ASR, media, subtitle, CLI, and VideoClipper workflows.
Configure GPT/Qwen/DeepSeek/AtlasCloud/MiniMax/LiteLLM/g4f/TwelveLabs Pegasus, fix API-key/base-url/model-prefix errors, or make LLM output parseable by AI Clip	llm-providers	Owns provider routing, prompt format, timestamp normalization, and no-live-call route smoke tests.
Build FunClip release archives, regenerate SHA256SUMS, align VERSION/README/release notes, inspect release workflow failures, or validate issue/PR templates	release-packaging	Owns versioned source archives, checksums, GitHub release contracts, and maintainer validation.

Minimal install and smoke check

From a FunClip checkout, install the declared runtime dependencies:

python -m pip install -r requirements.txt

For a no-network local smoke check after installation, run the bundled helper from the generated skill tree:

python scripts/check_environment.py --repo-root <funclip-checkout> --check-binaries

That helper verifies the requirements markers, imports core modules with the checkout's funclip/ directory on sys.path, checks launch kwargs, and uses a fake AutoModel to verify model selection without downloading weights.

Important boundaries

• Model weights are not bundled with FunClip. First real recognition may need network access or a pre-populated model cache.
• Live provider calls need provider-specific API keys and are not checked by the deterministic smoke scripts.
• Full video clipping and subtitle burn-in can require ffmpeg, ImageMagick, and a usable font.
• --listen binds all interfaces but does not create a public Gradio link; add --share only when that is intended.
• For programmatic use from a checkout, add the checkout's funclip/ directory to sys.path before importing launch, videoclipper, or llm.* helpers, because the source files use top-level imports.

Verification anchors

Useful native candidates for a current checkout include:

• tests/test_model_selection.py, tests/test_duplicate_text_matching.py, tests/test_recognition_result_compat.py, tests/test_gradio_runtime_compat.py, and tests/test_funasr_requirement.py for ASR/clip workflows.
• tests/test_openai_api.py, tests/test_minimax_api.py, tests/test_litellm_api.py, tests/test_twelvelabs_pegasus.py, and tests/test_minimax_launch_integration.py for LLM provider routing.
• tests/test_release_contract.py and tests/test_github_templates.py for release packaging and maintainer templates.

Run native tests only after choosing the relevant workflow and preparing the matching optional dependencies. The bundled smoke scripts are safer first checks because they avoid model downloads, credentials, and long media processing.