メニュー
Binary streaming between workers via channels. Use when building data pipelines, file transfers, streaming responses, or any pattern requiring binary data transfer between functions.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/ktg-one/content-hub --skill iii-channelsコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
SOC 職業分類に基づく
| name | iii-channels |
| description | Binary streaming between workers via channels. Use when building data pipelines, file transfers, streaming responses, or any pattern requiring binary data transfer between functions. |
Comparable to: Unix pipes, gRPC streaming, WebSocket data streams
Use the concepts below when they fit the task. Not every worker needs channels.
createChannel() returns a writer/reader pair plus serializable refs that can be passed to other workersreadAll(), and receive text messages via callbacksStreamChannelRef (e.g., ChannelReader::new(...)) rather than using the producer-side reader object returned by createChannel()A function creates a channel via createChannel(), receiving a writer and reader pair. The writer ref or reader ref is passed to another function (potentially in a different worker/language) via a trigger payload. The engine brokers the WebSocket connection between the two endpoints. Binary data flows directly between workers through the engine's channel endpoint.
| Primitive | Purpose |
|---|---|
createChannel(bufferSize?)| Create a channel, returns writer + reader pair |
ChannelWriter.write(data) | Send binary data (chunked into 64KB frames) |
ChannelWriter.sendMessage(msg) | Send a text message through the channel |
ChannelWriter.close() | Close the writer end |
ChannelReader.readAll() | Read entire stream into a single buffer |
ChannelReader.onMessage(callback) | Register callback for text messages |
StreamChannelRef | Serializable reference to pass between workers |
channel.writer.stream.write(buffer), channel.writer.sendMessage(text), channel.reader.stream, channel.reader.readAll(), channel.reader.onMessage(callback).iii.create_channel(), writer.write(bytes), writer.send_message(text), reader.read_all(), reader.on_message(callback).await iii.create_channel_async(), await writer.send_message_async(text), async iteration over reader.iii.create_channel(None).await, ChannelReader::new(engine_ws_base, &reader_ref), ChannelWriter::new(engine_ws_base, &writer_ref), next_binary(), read_all(), extract_channel_refs(&value).createChannel(), writer.sendBinary(uint8Array), writer.sendMessage(text), reader.onBinary(callback), reader.onMessage(callback), reader.readAll().Each reference shows the same patterns (channel creation, binary streaming, text messages, cross-function handoff) in its respective language.
Code using this pattern commonly includes, when relevant:
const channel = await iii.createChannel() — create a channel pair (producer access)channel.writer.stream.write(buffer) / channel.writer.write(data) — send binary datachannel.writer.sendMessage(JSON.stringify({ type: 'metadata', ... })) — send text metadatachannel.writer.close() — signal end of streamchannel.readerRef or channel.writerRef in trigger payloads for cross-worker streamingChannelReader::new(engine_ws_base, &reader_ref)const data = await reader.readAll() — read entire stream (consumer behavior)reader.onMessage(msg => { ... }) — handle text messages (consumer behavior)Use the adaptations below when they apply to the task.
readerRef to a processing function and writerRef to a producing function for pipeline patternsbufferSize when the reader may be slower than the writer to apply backpressureiii-state-management.iii-realtime-streams.subscribe type.iii-channels when the primary problem is binary data streaming between workers.iii-channels in the iii engine.KTG content marketing pipeline. Takes a written blog post and runs the full publishing workflow: repurpose for all platforms, generate hero image, SEO and AI citation optimisation, then pause for green-light before publishing to all channels via Composio. Local-first: uses Ollama for text tasks where quality allows, cloud for image gen and live data. Use when user says "hub", "/hub", "publish this", "run the pipeline", "post this everywhere".
KTG content marketing pipeline — SWARM MODE. Orchestrates 7 plugins via parallel agent dispatch. Drop a post, get it published everywhere with maximum parallelism. Local-first: Ollama for text, cloud for images and live data. Agents run simultaneously, not sequentially.
KTG content marketing pipeline — SWARM MODE v2. Concrete 11-agent parallel pipeline using the Agent tool. Local-first: Ollama for text, cloud for images and publish. Drop a post, spawn agents, publish everywhere.