Coding

算力超市skills

Try it

基于配置安全调用算力超市完整模型应用与任务开放接口

What it does

Plan and run intelligent workflows through a user-configured 算力超市 tenant. Use only when the user explicitly requests 算力超市, provides its configured tenant context, or invokes this skill for model, app, media, digital-human, music, smart-clip, asset, or social-media operations.

The skill document

算力超市skills

Operate 算力超市 as a capability graph, not as a collection of endpoints. Turn a user outcome into the smallest compatible workflow, obtain missing decisions only where they materially affect cost or output, then run and trace each node.

Credentials Guard

Before every network call, read persistent credentials from:

~/.likeadmin-market-skills/credentials.json

The user copies a JSON file to that exact location. Its only required fields are:

{
  "base_url": "https://your-tenant-api-origin",
  "api_key": "your-api-key"
}

For installation surfaces that support copyable environment variables, provide this setup block. These variables are only used by the user's shell to populate the local credential file; the skill does not read them at runtime.

export LIKEADMIN_MARKET_BASE_URL="https://your-tenant-api-origin"
export LIKEADMIN_MARKET_API_KEY="your-api-key"
printf '%s\n%s\n' "$LIKEADMIN_MARKET_BASE_URL" "$LIKEADMIN_MARKET_API_KEY" | node scripts/api-client.mjs config import-lines
unset LIKEADMIN_MARKET_BASE_URL LIKEADMIN_MARKET_API_KEY
node scripts/api-client.mjs config status
  • base_url is a tenant API origin. Remove a trailing /; never append or hardcode a tenant domain.
  • Send api_key only as Authorization: Bearer . Never print it, place it in URLs, or write it into workflow state.
  • node scripts/api-client.mjs config path prints the target location. config status reports presence only. The helper creates the directory with 0700 permissions and the credential file with 0600 permissions.
  • manifest.yaml is the machine-readable declaration of this boundary and labels the two copyable variables as setup-only. Do not call any API when either field is absent.

Plan Before Calling

  1. Classify the request as a direct capability, a multi-step outcome, an inspection/query, or a destructive asset operation.
  2. Discover the tenant's actual availability before selection: GET /models?detail=1, GET /apps, and GET /apps/{app_code}. Use GET /pricing or /pricing/batch before a cost-sensitive task.
  3. Read capability-catalog.json for type contracts, then use node scripts/read-contract.mjs for the matching full contract. The bundled snapshot covers all 136 published documents, including 29 model and 72 application contracts, but is not proof that a resource is enabled for this tenant.
  4. Build a DAG only where an upstream output satisfies a downstream input type. Preserve an explicit model/app choice. If several eligible alternatives differ materially in price, quality, duration, audio, or resolution, ask the user to choose. Never silently choose a premium option.
  5. Validate every required field and constraint before spending points. Confirm destructive asset deletion. Do not retry voice cloning automatically because a retry can create another voice.

Media URL Boundary

Every local path, attachment, file://, blob:, or private/unreachable URL used in a business URL field must first go through:

POST ${base_url}/api/v1/upload
multipart field: file

Require code=1 and use only data.url in the dependent request. Do not upload already-public http(s) URLs. These input URLs are temporary, normally at most 24 hours, so submit dependent nodes immediately. storage is output transfer, never an input uploader.

Before uploading any local path, explicitly tell the user the configured tenant API destination and obtain confirmation for that specific transfer. The workflow runner enforces confirm_local_upload: true for local files.

Use the catalog's media_url_fields as the baseline, then inspect the selected contract for nested fields such as content[].image_url.url, media[].url, materials[].fileUrl, and garment.data[].url.

Execute Nodes Correctly

  • Text models: use POST /chat/completions; stream=true is SSE, not a pollable task.
  • Async model generation: use POST /tasks with model and business fields at the JSON root. Do not mix root business fields with a non-empty legacy params object.
  • Application APIs: use the exact method and ${base_url}/api/v1/apps/{app_code}/{api_code}. Preserve case-sensitive fields such as Seedance URL, Name, GroupId, and ProjectName.
  • Do not force an application mode=async or mode=task. Some applications use mode as their own quality or generation setting.
  • Treat {code:0} as failure even when HTTP status is 200.
  • For a platform task_id, poll GET /tasks/{task_id} until completed, failed, or cancelled. Use an app-specific query only when its contract requires an upstream/elastic ID or the user specifically requests it.

Use an explicit callback_url only when the user supplies a reachable endpoint. Callbacks lack a signature header, so retain polling as recovery.

Deterministic Workflow Runner

Use workflow-runner.mjs for fragile, repeatable chains after the agent has made the required capability and template choices. It uploads local media, verifies each platform task before continuing, and can persist non-secret node state with --state .

Supported recipes include clone-voice, tts, lipsync, clone-voice-lipsync, image-avatar, realman-smart-clip, avatar-smart-clip, material-smart-clip, and news-smart-clip. Read workflow-inputs.md before running one.

The runner intentionally does not select a Smart Clip style_id, force a quality/cost option, delete an asset, or cross application-specific IDs. Those remain user-confirmed orchestration decisions.

Sensitive Operations

Use social-media lookup, search, listing, and watermark-removal APIs only for content the user owns or is authorized to process. Do not bulk enumerate profiles or combine search, listing, detail retrieval, and watermark removal into a collection workflow. For a destructive asset delete or sensitive media operation, obtain explicit confirmation immediately before the request and communicate in the user's language.

Composition Rules

Read orchestration.md for supported recipes and non-combinable IDs. The important boundaries are:

  • A voice_tts clone ID is valid for voice_tts TTS, not automatically for music-generation persona/voice fields.
  • lipsync requires both an audio URL and a source video URL; a still image is not a valid substitute.
  • Smart Clip always starts with a scene-compatible template. realMan, oralMixCutting, and newsMixCutting are not interchangeable.
  • Only pass a generated result forward after confirming it is a reachable URL of the target media type. IDs such as audio_id, persona_id, asset_id, and template IDs stay within the applications that define them.
  • If no enabled capability matches, explain the missing prerequisite or unavailable resource rather than fabricating a workflow.

Reference Routing

  • capability-catalog.json: complete online snapshot index, live-at-snapshot flag, schemas, URL fields, billing and operation semantics.
  • orchestration.md: intent routing, DAG recipes, branching decisions, and safety constraints.
  • contracts.bundle.json: full local contract bundle; use read-contract.mjs to retrieve a single document without loading all of it.
  • api-client.mjs: persistent-credential guard, discovery, upload, request, and polling helper.

Related skills

This skill should be used when the user needs strategic advice on AI-driven business transformation, including digital marketing automation, AIGC commercial applications, AI agent design, smart commerce, or business intelligence. Triggers on keywords like: 数字营销, AIGC, AI智能体, 智慧商务, 商务智能, AI转型, 营销自动化, AI agent, 商业AI.

2 installs

Agent skill recommender. Input a user need, task description, or existing skill list; output best matching skills, install rationale, duplicate/merge candida...

37 installs

Search, discover, browse and install AI Gallery Agent skills via natural language. Triggers include: "AI Gallery", "AI Gallery有什么skill", "有什么skill", "AI Gallery相关skill", "AI Gallery agent skill 市场", "AI Gallery skill类目", "skill 市场", "搜索AI Gallery skill", "安装skill", "订阅skill", "有没有XX skill", "有没有XX的能力", "帮我找 XX skill", "帮我找一个能XX的工具", "我想扩展功能", "介绍 XX Skill 内容", "XX Skill 具体做什么", "explore AI Gallery skills", "show AI Gallery skill categories", "does an AI Gallery skill exist for...", "which AI Gallery skills exist", "search skill", "find skill".

1 installs

Curated catalog of quality AI agent skills. Find the right tool for every task — verified recommendations for memory, productivity, security, and automation....