thumbnail-skill
Deterministic thumbnail rendering execution Skill for the AI Video Production Ecosystem. Use it when a caller (normally video-production-agent) has already decided exactly what a thumbnail should show - which still image or which explicit video timestamp, what text, what layout - and needs it rendered safely to PNG/JPEG with provenance. Do NOT use it to choose what a thumbnail should show, pick a "best" video frame, detect faces, generate a title, judge design quality, or run arbitrary ffmpeg/image commands (it refuses them); those decisions belong to the caller.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- thumbnail-skill
- description
- Deterministic thumbnail rendering execution Skill for the AI Video Production Ecosystem. Use it when a caller (normally video-production-agent) has already decided exactly what a thumbnail should show - which still image or which explicit video timestamp, what text, what layout - and needs it rendered safely to PNG/JPEG with provenance. Do NOT use it to choose what a thumbnail should show, pick a "best" video frame, detect faces, generate a title, judge design quality, or run arbitrary ffmpeg/image commands (it refuses them); those decisions belong to the caller.
02 资源文件 43
- 参考文档.github/FUNDING.yml17 B
- 参考文档.github/workflows/tests.yml2.9 KB
- 其他.gitignore98 B
- 资产文件brand/app-icon-dark.png56.5 KB
- 资产文件brand/app-icon.png56.5 KB
- 资产文件brand/logo/thumbnail-skill-wordmark-black.svg4.0 KB
- 资产文件brand/logo/thumbnail-skill-wordmark-white.svg4.0 KB
- 资产文件brand/thumbnail-skill.icon/Assets/frame.svg319 B
- 资产文件brand/thumbnail-skill.icon/Assets/scene.svg414 B
- 参考文档brand/thumbnail-skill.icon/icon.json1.7 KB
- 参考文档docs/architecture.md5.2 KB
- 参考文档docs/decisions.md5.8 KB
- 参考文档docs/ffmpeg-skill.md6.6 KB
- 参考文档docs/model.md4.8 KB
- 参考文档docs/security.md5.4 KB
- 参考文档docs/testing.md3.7 KB
- 许可证LICENSE1.0 KB
- 其他pyproject.toml868 B
- 参考文档README.md17.5 KB
- 参考文档skill.contract.json9.6 KB
- 脚本src/thumbnail_skill/__init__.py1.0 KB
- 脚本src/thumbnail_skill/adapter.py21.1 KB
- 脚本src/thumbnail_skill/canonical.py917 B
- 脚本src/thumbnail_skill/cli.py10.7 KB
- 脚本src/thumbnail_skill/contract.py10.5 KB
- 脚本src/thumbnail_skill/doctor.py3.9 KB
- 脚本src/thumbnail_skill/errors.py2.8 KB
- 脚本src/thumbnail_skill/executor.py24.7 KB
- 脚本src/thumbnail_skill/fonts.py6.4 KB
- 脚本src/thumbnail_skill/model.py26.9 KB
- 脚本src/thumbnail_skill/renderer.py7.7 KB
- 脚本src/thumbnail_skill/security.py8.0 KB
- 脚本src/thumbnail_skill/skill.py2.3 KB
- 脚本tests/__init__.py—
- 脚本tests/conftest.py1.9 KB
- 脚本tests/fixtures/__init__.py—
- 脚本tests/fixtures/generate.py2.4 KB
- 脚本tests/test_cli.py6.8 KB
- 脚本tests/test_contract.py2.3 KB
- 脚本tests/test_ffmpeg_skill_compat.py9.3 KB
- 脚本tests/test_integration.py21.5 KB
- 脚本tests/test_security.py7.8 KB
- 脚本tests/test_unit.py9.0 KB
03 SKILL.md 正文
thumbnail-skill
Machine interface: thumbnail run - --json with a {"tool": "thumbnail/<name>", "params": {...}}
document on stdin; exactly one response document on stdout. thumbnail skill --json prints the
contract, thumbnail doctor --json the environment (fonts, ffmpeg-skill, path policy).
Tools: thumbnail/validate (structural check only), thumbnail/render (canvas + assets + elements
-> PNG/JPEG), thumbnail/extract_frame (one video timestamp -> one frame, nothing else).
Rules for a calling agent:
- Decide everything about content before calling this skill: which still image, or which exact
video timestamp (
asset.timestamp, seconds); what the text says (with literal\nfor line breaks — this skill never wraps or reflows text); where every element sits (position,size,z_index). This skill renders exactly what it is given and searches for nothing. - Reference fonts by
font_idfrom the registry (thumbnail skill --json->fonts.font_ids, orthumbnail doctor --jsonfor what actually resolves on this machine). A request never carries a font path; an unresolvablefont_idfailsMISSING_INPUT, never a silent substitution. - Never send
command,argv,shell,filter/filter_complex/vf/af,executable,env, or HTML/CSS/JS: the request is rejected withINVALID_REQUESTwherever such a key appears. - Keep outputs inside the workspace (
--workspace/options.workspace), never at an input path; setoverwrite: truedeliberately. Restrict inputs withoptions.allowed_input_rootswhen the caller's own inputs come from an untrusted or shared location. - Read
output,sha256,width/height,reusedandprovenancefrom the response; the exit code is 0 only whenokis true. Avideo_frameasset needs ffmpeg-skill reachable (env varTHUMBNAIL_SKILL_FFMPEG_SKILL_DIRor--ffmpeg-skill); a still-image-only document never needs it.
See README.md for the ThumbnailDocument schema, element/font/output tables, error codes, security boundary and the responsibility split with video-production-agent and the other media skills.
04 同仓库更多 Skills
video-editing-skill
Deterministic video editing from a typed edit request (trim, cut, concat with transitions, speed change, fit / fill / resize, still-image overlay) with an exact source-to-timeline mapping and per-operation provenance. Executes through ffmpeg-skill; never takes commands, argv, filter strings or executables. Use when an agent has already decided what to edit and needs the edit performed, verified and traced. Not for deciding what to cut, transcribing, captioning, colour grading or audio mastering.
subtitle-skill
Deterministic subtitle validation, SRT/WebVTT generation, and video caption burn-in. Given an already-decided, typed subtitle document (cue text, timing, optional speaker/style), validate its timeline, generate an SRT or WebVTT file, or burn it into a video by delegating to ffmpeg-skill's caption tool. Use this skill only after another agent has already decided what the subtitle cues should say and how they should be split -- it performs no transcription, no diarization, no cue-splitting judgement, and no "does this look right" visual check.
ffmpeg-skill
Edit video and audio with local FFmpeg from natural-language requests: cut, trim, join, resize/reframe (9:16, 1:1), speed change, captions and subtitles (SRT/ASS, animated, karaoke), logos and text overlays, lower-thirds and titles, silence removal, multicam and external-mic sync, loudness normalisation, HDR/Dolby Vision to SDR, LUTs, background music with ducking, platform exports (YouTube, Reels, TikTok, X), compliance checks, scene detection and highlight reels, contact sheets to inspect results, and whole-edit project files. Use this skill whenever the user mentions a video or audio file (mp4, mov, mkv, wav, m4a), footage, a clip, captions, subtitles, a reel or short, YouTube/Instagram/TikTok delivery, LUFS, sync, transcoding, ffmpeg, or asks to make something "60 seconds", "vertical", "louder", "captioned" — even when they do not say "edit". Python 3.9 standard library only, no cloud, no API keys.
skillpact
Write and check SkillPact contracts (skill.contract.json) for Agent Skills with the skillpact CLI - draft a contract from a SKILL.md, validate it, run its conformance tests network-isolated, hash and sign every file, report the level L1 / L2 / L3, match it against a harness policy (allow / ask / deny) and compile that decision to Claude Code, Cursor, Codex or Gemini CLI settings or export it to HushSpec, Rego, APS or ACS. Use when the user wants a skill's permissions declared or reviewed before install, a contract drafted, validated, signed or migrated, a conformance level or badge, a policy check, or harness permission settings made from a policy. Python 3.11+; bubblewrap on Linux for L2 and L3. Not a runtime sandbox and not a malware scanner - L3 means these files, signed by a trusted author, passed their tests, not that the skill is safe.