add-byok-provider
Skill10k starsChanged 4 days agoApache-2.0View on GitHub
Add or maintain a built-in BYOK media provider in the OpenMotion repository, including official API research, credentials, model discovery, generation adapters, editor configuration, provider references and verification. Use for contributor requests to integrate a supplier; ordinary media generation uses the existing production guidance.
1 other project uses a version of this.
Trending
More talk lately.about the whole repository
Most recently taken up by
- nexu-io/openmotionword for word, first seen Nov 15, 2026
Who's talking about nexu-io/html-anything
- Prism Labs“html-anything — One Editor, 8 AI Agents, Every Format You Ship”YouTube · May 15, 2026 · 767 views
- 思思主播“不是 Markdown vs HTML:用 51 毫秒做出 agent 三層交付”YouTube · May 19, 2026 · 91 views
- さつきのOSS研究室“【html-anything】Markdown 10 本を Claude Code で雑誌風 HTML にしたら表とコードは全部残り、見出しと冒頭は 15 か所書き換わった”YouTube · Nov 11, 2026 · 25 views
- BarbuPrime IA | Claude Code Vs Codex“3 diseños que Claude Code hace con tus notas 🎨 HTML Anything (sin API key)”YouTube · Oct 25, 2026 · 10 views
- nexu-io/openmotion“feat(agents): add ZCode (Z.AI) adapter”GitHub · Aug 12, 2026 · 3 reactions
The biggest 5 of 23 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 23
Talked about on GitHub, YouTube, Bluesky and Hacker News (23 times).
On GitHub · 16
- Aug 12, 2026feat(agents): add ZCode (Z.AI) adapter — nexu-io/openmotion3 reactions · 5 comments
- Nov 9, 2026feat(motion): ship complete npm application and verified Agent installation guide — nexu-io/openmotion2 reactions · 1 comment
- Nov 11, 2026feat(studio): refine Motion MVP editing, feedback and empty states — nexu-io/openmotion1 reaction · 2 comments
- May 18, 2026fix: Qwen Coder responses not rendering — enhance stream-json parser — nexu-io/openmotion1 reaction · 6 comments
- May 17, 2026chore(featured): curate Featured picks down to top 10 across categories — nexu-io/open-design1 reaction · 1 comment
11 more on GitHub
- Nov 10, 2026agent-skills.github.nexu-io.html-anything: 0-unstable-2026-10-13 → 0-unstable-2026-11-10 — sudosubin/agents.nix0 reactions
- Nov 3, 2026[Skill Proposal]: html-anything — akhmadilham/pinax-registry0 reactions · 1 comment
- Oct 14, 2026agent-skills.github.nexu-io.html-anything: 0-unstable-2026-09-15 → 0-unstable-2026-10-13 — sudosubin/agents.nix0 reactions
- Oct 13, 2026Embed html-anything skill as Morph AI's base skill for HTML output — FloatLikeABee/robo0 reactions
- Oct 3, 2026agent-skills.github.nexu-io.html-anything: init at 0-unstable-2026-09-15 — sudosubin/agents.nix0 reactions
- Sep 30, 2026[Candidate] nexu-io/html-anything - large collection (81 SKILL.md files) — ProSkillsMD/proskills0 reactions · 1 comment
- Aug 5, 2026idea: nexu-io/html-anything — zqf1314/idea0 reactions · 1 comment
- May 18, 2026Audit candidate: nexu-io/html-anything — xiaolai/nlpm0 reactions · 5 comments
- May 15, 2026docs: add HTML Anything visual page — 53able/docs-site0 reactions
- May 15, 2026TrendForge Daily — 2026-05-15 — anirudhDVDEBUG/RnD0 reactions · 1 comment
- May 13, 2026docs: fix placeholder URLs and refresh PROGRESS for v5 — nexu-io/openmotion0 reactions
On YouTube · 4
- May 15, 2026html-anything — One Editor, 8 AI Agents, Every Format You Ship — Prism Labs767 views · 1 comment
- May 19, 2026不是 Markdown vs HTML:用 51 毫秒做出 agent 三層交付 — 思思主播91 views · 1 comment
- Nov 11, 2026【html-anything】Markdown 10 本を Claude Code で雑誌風 HTML にしたら表とコードは全部残り、見出しと冒頭は 15 か所書き換わった — さつきのOSS研究室25 views · 1 comment
- Oct 25, 20263 diseños que Claude Code hace con tus notas 🎨 HTML Anything (sin API key) — BarbuPrime IA | Claude Code Vs Codex10 views · 1 comment
On Bluesky · 2
- Sep 2, 2026Markdown 只是草稿 给人看的成品应该直接是 HTML HTML Anything 干的就是这步 笔记、CSV、Excel、JSON 丢进去 它调用你电脑上已经登录的 coding agent 吐一份能发布的单文件 HTML 不用再填一套 API Key Claude… — @anotherbug.com1 likes and repost
- Aug 2, 2026HTML Anything ships no model and no API key. It detects a coding-agent CLI you already logged into and drives it to turn notes or Markdown… — @agentpalisade.bsky.social1 likes and repost
On Hacker News · 1
- May 20, 2026HTML-anything – The agentic HTML editor – your local AI agent writes the HTML3 points
Other versions1 project uses a version of this.
Word for word · 1
- nexu-io/openmotion★ 10k100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- openmotion-directorSkill
- openmotion-productionSkill
- openmotion-recreateSkill
- openmotion-reference-videoSkill
- openmotion-template-to-videoSkill
- openmotion-video-to-templateSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 43
Use the exact model schema.
- Distinctive: line 44
Keep fixed provider origins where possible.
- Very specific: line 50
Create skills/openmotion-production/references/media-providers/<provider-id>.md and register it in the production workflow.
- Distinctive: line 58
Use deterministic fixtures to cover the new provider's real boundaries:
≈ “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
Add a BYOK provider
Use this repository-local contributor Skill to implement a provider through the existing openmotion_media_use Tool and Model Gateway. Read the checkout's AGENTS.md and follow its current development/PR rules. Paths below are relative to the repository root. This Skill stays in .agents/skills/; product guidance ships from root skills/.
Start with gateway behavior, then inspect the current callers and the closest existing adapter. Use Atlas for a shared image/video/audio prediction API, fal for returned queue URLs, WaveSpeed for authenticated discovery, and OpenDesign for a custom HTTPS gateway. Check current code instead of copying an entire provider or its assumptions.
Establish the provider contract
Find and read the provider's official API/model documentation. Record source links and the date checked in its reference. Establish:
- Stable provider ID, display name, key-creation URL, API origin and authentication scheme.
- Whether model discovery is public or authenticated, response fields, pagination and model-specific schema links.
- Supported image/video/audio operations, exact model IDs, required fields and sample outputs. Separate speech from music/SFX and media files from transcription, text, training or 3D results.
- Submit and poll endpoints, remote task identity, intermediate/terminal statuses, output retrieval, cancellation and idempotency semantics.
- HTTP errors versus envelope/task codes, rate limits and uncertain outcomes. A 401 alone does not prove an incorrect key; consult the provider's actual semantics.
Implement only confirmed API behavior. Do not invent a model catalogue endpoint, account entitlement, unsupported modality, remote cancellation or refund. If documentation leaves a blocking ambiguity, explain the missing contract while continuing independent work.
Extend the existing integration
Inspect these boundaries and update the ones the provider needs:
| Boundary | Current source | Required result |
| Provider identity | shared/gateway.ts, gateway/catalog.ts | Add the provider ID and label; keep shared types usable by Studio and MCP. |
| Credentials and availability | gateway/service.ts | Resolve the provider's environment key and service-owned saved key, retaining saved-key priority and existing save/remove semantics. A configured key enables the provider; it does not prove successful generation. |
| Model discovery | gateway/catalog.ts, gateway/service.ts, optionally gateway/providers.ts | Query the documented catalogue with the correct authentication. Normalize only supported media outputs; preserve pagination, caching and unavailable-catalogue warnings. Built-in fallback entries are optional examples, not a replacement for live discovery. |
| Generation | gateway/providers.ts, or a focused provider module | Implement ProviderAdapter.generate and register it in providerAdapters. Reuse persisted jobs, progress/remoteId, validation, downloads and project import. |
| Editor configuration | studio/src/motion/ModelSettings.tsx | Include the provider card with the correct key-setup/usage links. Do not let a new provider fall through to another supplier's key URL. Integrate catalogue sync, capability filtering and defaults in the settings center. Keep account authorization separate from provider API keys. Distinguish listed presets from implemented execution protocols. |
| Reference delivery | shared/workflow.ts, server.ts, gateway/media-use.ts | Register the provider document under production resources and include its HTTP reference route. Confirm inspect/reference/models accept and return the new ID. |
| User guidance | skills/openmotion-production/references/media-providers/, references/materials.md in that production Skill, relevant docs/ | Maintain one provider reference, linked from shared material guidance and available through the Tool. |
The public Tool already covers inspect/reference/models/generate/status/cancel. Reuse those actions; a supplier does not need its own MCP Tool or distributable Skill. Preserve the existing production routes and intentional UI visibility decisions. Tool and Skill lists come from their actual registries; do not introduce fixed quantity gates.
Request and credential boundaries
- Inject keys only inside the service. Never read a contributor's real secret file, return a key to an Agent, add credentials to model parameters, or commit them. Use synthetic credentials in fixtures.
- Use the exact model schema. Some speech models expect
text, othersprompt; the existing mode controls that mapping. Cloud speech additionally nests options. Check the wire body, not just the Tool arguments. - Keep fixed provider origins where possible. Any user-configurable origin must retain existing public HTTPS/DNS validation. Validate returned task IDs/queue URLs, and never forward authorization to an unrelated output or redirect host.
- Save the remote ID before polling and distinguish accepted, processing, failed and completed outputs. Do not automatically retry a paid POST, switch paid suppliers, or create a new task after a timeout. Check the original job/history first. Local cancellation/timeouts do not prove a remote stop.
- Reuse
apiJson, media validation and project-owned imports. Recognize their limits: current non-2xx handling omits raw provider bodies/headers, and local errors can omit task detail. Do not promise full provider diagnostics or infer a definitive cause from generic local wording.
Write the provider reference
Create skills/openmotion-production/references/media-providers/<provider-id>.md and register it in the production workflow. Read one existing document, such as Atlas, for the maintained shape; do not duplicate its supplier-specific facts.
Include key setup, authentication and model-list queries; Tool JSON templates for each supported modality; their exact HTTP endpoint/body mapping; polling/result retrieval; representative error codes with meaning, next action and retry boundaries; official source/model links and known integration limits. Distinguish Tool arguments from wire JSON, local job IDs from remote task IDs, and HTTP status from application/task codes. Mark unspecified behavior as unknown.
Keep real keys and internal/private source material out of the document. Show environment placeholders in wire examples, and direct Agents through openmotion_media_use so the service owns credentials. Templates do not execute paid requests by themselves. Do not add this contributor Skill to packagedSkills, workflow resources or host installation lists.
Verify and hand over
Use deterministic fixtures to cover the new provider's real boundaries:
- Model discovery/authentication and normalization, including exclusion of unsupported output types.
- Public media Tool -> exact provider URL/auth/body -> original task polling -> validated project-owned file, for each implemented modality. Assert that an idempotent replay submits once.
- Rejected authentication/input and unsafe task/output destinations; errors or cancellation must not trigger another paid submission or expose a key.
- Configuration save/clear/remove, correct setup/reference links, catalogue refresh/error recovery, and default model selection.
- Reference delivery through Tool, MCP resources and HTTP, plus inclusion of the reference in distribution artifacts.
Adapt existing tests in tests/byok.test.ts, studio/src/motion/ModelSettings.test.tsx, tests/workflow.test.ts, tests/npm-package.test.ts and tests/byok-browser.mjs. Add assertions for observable behavior, not documentation wording. Ensure a new test file is included in the repository's explicit CI runner.
For runtime/Studio changes run npm run check, npm run typecheck:studio, npm run build:studio and affected deterministic/packaging checks under the pinned Node version. Use the shared browser-lock lane for real browser tests when available, or the exclusive isolated lane in tests/byok-browser.mjs; always use an isolated test project/data directory. Follow AGENTS.md when selecting narrower checks for documentation-only work.
Review docs/product/architecture.md and update it only if product behavior or boundaries changed. Report implemented modalities, references, actual checks, untested live-provider behavior and remaining limits. Run real paid generation only within the user's requested scope; fixture success is not account/billing or audiovisual acceptance. Commit/push/PR/merge/publication follow the current task authorization separately.
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.

