polym

Research BytePlus products, APIs, SDKs, setup, quotas, regions, pricing, and troubleshooting using the bundled official documentation index, then verify claims on live docs. Use for requests for BytePlus documentation or official links.

Batch-generate Seedance 2.0 Mini videos via the BytePlus console BFF (Mini has no public API). Covers T2V, image/video reference (I2V/R2V), first/last frame, and human portraits. Triggers: "seedance mini", "dreamina mini", "用 mini 出视频".

Generate textured 3D models (GLB + PBR) from an image and/or text prompt via Volcengine Ark Seed3D (doubao-seed3d). Submits the async task, polls to completion, and downloads the .glb. Domestic Volcengine Ark only, not BytePlus overseas.

Generate videos with BytePlus Seedance. Use for Seedance text-to-video, image-to-video, camera control, and draft video workflows.

Generate images with BytePlus Seedream 4.x/4.5. Use for Seedream text-to-image, AI art, product images, and batch image generation.

Creates polished explainer videos from docs, URLs, PDFs, screenshots, or text. Use for product demos, announcement reels, repo videos, and requests like "make an explainer video" or "turn this repo into a video".

Generate Slack daily digests, highlight customer risks and action items, and publish summaries to Lark or local outputs.

Score and qualify Trigify leads using LLM-powered evidence extraction and deterministic routing. Supports Anthropic, OpenAI, Gemini, BytePlus Ark, and any LiteLLM-compatible provider.

Fetches yesterday's Lark IM messages, organises topics and key content, materialises a Feishu digest doc for the day, and saves automatically to the knowledge base. Trigger phrases: 'organize yesterday's messages', 'daily IM digest', 'pull yesterday's group messages', 'im digest'.

Generate images with gpt-image-2 (ChatGPT-image-2) via the internal AutoCase playground (autocase.bytedance.net/arena), supporting reference images for image-to-image. Use when the user wants gpt-image-2 / ChatGPT-image-2 — text-to-image, or image-to-image driven by reference photos. Trigger on "generate with gpt-image-2", "use autocase playground", "ChatGPT-image-2 一张图", "用 autocase 出图". Drives a real Chrome browser, shows the result inline, and saves it to ~/Downloads (or a folder the user specifies).

Generate Gemini/Imagen images via Vertex AI, AI Studio, or AIDP. Use for Gemini image generation, editing, and reference-image workflows.

Generate or edit images with GPT Image via ByteDance AIDP. Use for gpt-image-1 text-to-image, multi-image edits, and masks.

Generate videos with Sora via OpenAI or ByteDance AIDP. Use for Sora 2/Pro text-to-video and image-to-video workflows.

Generate videos with Google Veo via Vertex AI or AIDP. Use for Veo text-to-video, image-to-video, and first-frame workflows.

Optimize prompts for Seedream, Seedance, Gemini, Sora, and Veo. Use when the user wants a stronger image/video generation prompt.

Score generated images with Gemini via Vertex AI or AIDP. Use for pointwise image grading and pairwise A/B/tie image comparisons.

Score generated videos with Gemini via Vertex AI or AIDP. Use for pointwise video grading and pairwise video comparisons.

SQLite eval database for image/video model comparisons. Use to import results, query win rates, dimensions, examples, and Elo rankings.

Generate Markdown reports from polym-eval-db. Use for eval summaries, model comparison reports, Elo reports, and exported evaluation findings.

Run model generation, pairwise Gemini scoring, and eval-db persistence. Use for new image/video benchmarks, GSB runs, and model comparisons.

Seed LLM GTM and competitive positioning. Use for Seed 1.8/2.0 strengths, gaps, benchmarks, pricing, and customer messaging.

Seedance video GTM and competitive positioning. Use for Seedance strengths, gaps, Sora/Veo/Kling comparisons, and customer messaging.

Seedream image GTM and competitive positioning. Use for Seedream strengths, gaps, Gemini/Nano Banana comparisons, and messaging.

Query the C360 data dashboard by customer name and specific request, presenting usage trends and insights. Trigger phrases: '查一下客户X的数据', 'dashboard watch', 'X used how many tokens', '看一下X的用量'.

Pre-meeting customer intelligence snapshot: input a customer name and get a 2-minute briefing in ~30 seconds. Trigger phrases: '帮我备一下 X 的课', 'customer brief X', 'X customer intelligence', 'prep for meeting with X'.

Onboarding guide for Polym — what it does, how to install it, common use cases, and how to contribute new skills. Trigger when a user seems lost, asks what this is, or asks how to get started.

Scan newly added/modified Lark documents over a time period, automatically extract and ingest into the knowledge base. Default range: yesterday 00:00 to now. Supports ingesting a single specified document. Trigger phrases: '把最近的文档入库', 'lark doc ingest', 'ingest recent docs', 'save this document to knowledge base'.

Initialize a local LLM-maintained wiki for an SA. Creates the full directory structure, SCHEMA.md, PROMPTS.md, and starter wiki pages. Invoke with: 'set up my local wiki', 'init local wiki', '初始化本地 wiki', 'create my SA wiki'.

Live meeting recording for SA work. Uses Lark Calendar to identify the current meeting, starts Syncore note-taker recording, supports mid-meeting Q&A/progress summaries, and saves the final AI summary to the knowledge base. Trigger phrases: '开始录会', '开始会议录制', 'record this meeting', 'start meeting notes'.

Compile all meetings (Lark VC + Minutes) over a time period, extract customer insights and key conclusions, and automatically save to the knowledge base. Default range: yesterday 00:00 to now. Trigger phrases: '整理最近的会议', 'meeting summary', '帮我总结一下昨天的会', '会议纪要整理'.

BytePlus SA Team knowledge base — read and write. Use when asking about SA knowledge, customer profiles, meeting notes, archiving information, checking customer progress, adding topic pages, or ingesting handbook chapters. READ path: Bitable knowledge_index search + full-text docs fetch. WRITE path: mandatory write_queue (agents never write wiki pages directly). Multi-agent concurrency-safe. Operational tutorials (containing URL + step-by-step instructions) trigger an optional Chrome MCP browser demo offer — see references/browser-demo.md.

One sentence, ≤ 250 chars, starting with a verb. Describe what the skill does AND when Claude should pick it. Include 2-3 trigger phrases inline.