Byted Mediakit Shared

MediaKit Shared Rules

This skill explains how to operate on media resources via mediakit-cli, and the shared rules and caveats that apply to every call.

Prerequisites

Install the CLI

Before first use, confirm the CLI is installed:

# Install
npm install -g @volcengine/mediakit-cli

# Verify
mediakit-cli --version

Authentication check

Priority: environment variables > config file (path: ~/.mediakit/config.json).

Field reference

• Environment variables / config file: MEDIAKIT_API_KEY, MEDIAKIT_ENDPOINT, MEDIAKIT_SURFACE, MEDIAKIT_RUNTIME

When any required field is missing, stop execution and output the full list of missing items together with fix suggestions.

Cloud calls automatically carry x-surface / x-runtime. Header priority: environment variables > ~/.mediakit/config.json > defaults / environment probing. When this Skill/Plugin invokes cloud capabilities through mediakit-cli, the runtime should inject MEDIAKIT_SURFACE=skill|plugin and MEDIAKIT_RUNTIME=. The CLI keeps the original product prefix and reports x-surface=cli/skill or x-surface=cli/plugin. If no explicit configuration is provided, the CLI defaults to x-surface=cli; x-runtime falls back to IDENTITY_NAME / OPENCLAW_SERVICE_MARKER environment probing, and finally to unknown.

Source-reporting constraints

• When a Skill invokes mediakit-cli, it must explicitly set MEDIAKIT_SURFACE=skill and must not rely on the user's pre-existing environment variables.
• When a Plugin invokes mediakit-cli, it must explicitly set MEDIAKIT_SURFACE=plugin and must not reuse the Skill value.
• The host runtime identifier should also be explicitly set via MEDIAKIT_RUNTIME=; if unset, the CLI falls back to environment probing or unknown.

MEDIAKIT_SURFACE=skill MEDIAKIT_RUNTIME= mediakit-cli editing add-image-to-video

MEDIAKIT_SURFACE=plugin MEDIAKIT_RUNTIME= mediakit-cli editing add-image-to-video

CLI usage

Initialization

Run the init wizard before first use:

mediakit-cli init

For non-interactive initialization (Agents), pass the request-source and runtime configuration explicitly:

mediakit-cli init --mode cloud-first --api-key  --runtime  --surface cli --yes
mediakit-cli init --mode local-first --api-key  --endpoint  --output-path ~/mediakit-output --runtime  --surface cli --credential-store config --yes

Common commands after initialization:

# Show current configuration
mediakit-cli config show

# Switch the default mode to local-first
mediakit-cli config set mode local-first

# Switch the default mode to cloud-first
mediakit-cli config set mode cloud-first

# Refresh environment checks and view dependency status
mediakit-cli doctor

Command structure

MediaKit CLI consistently uses the domain + tool invocation shape:

mediakit-cli {domain} {tool} [flags]

Common help commands:

# List all domains
mediakit-cli --domains

# List tools in a given domain
mediakit-cli {domain} --help

# Show parameters of a specific tool
mediakit-cli {domain} {tool} --help

# Dynamically discover tool capability and return structure
mediakit-cli {domain} {tool} --schema
mediakit-cli --local {domain} {tool} --schema

Domains currently covered by this product: editing, video.

Schema discovery

Every capability command supports --schema, used by Agents to dynamically read tool capabilities. Required business parameters are not enforced when --schema is used.

The returned structure contains:

• name: tool name in snake_case, e.g. add_image_to_video
• description: tool description, automatically including Mode and Async information
• input_schema: input parameter JSON Schema
• output_schema: return structure for the current execution mode

Output disambiguation rules:

• By default the return surface is resolved against the global mode configuration
• --local ... --schema outputs the local-mode return surface; local mode returns the final result fields directly
• Cloud async tools output task_id / request_id, and describe the completed-state result of query-task under final_result
• query-task is cloud only; its schema describes task status plus the completed-state result

Examples:

mediakit-cli editing trim-video --schema
mediakit-cli --local editing trim-video --schema

Per-invocation mode override

In addition to config set mode for setting the default mode, you can override the mode for a single command only:

mediakit-cli --local editing add-image-to-video

mediakit-cli --cloud editing add-image-to-video

Additional rules:

• --local / --cloud only affect the current command and do not modify the global config.mode
• --local and --cloud are mutually exclusive and cannot be passed together

Async tasks

When an async media-processing task is accepted successfully, the response contains a task_id field. Use the shared query-task command to poll the result.

mediakit-cli shared query-task --task-id 

local / cloud constraints

• query-task is a cloud only tool
• local mode does not support query-task
• The current capability set runs primarily in cloud mode; when an explicit declaration is required, prefer --cloud

Cloud-mode media input notes

• When a command runs with --cloud or under the cloud-first strategy, media input parameters (video_url, audio_url, image_url, subtitle_url, sub_image_url, and their corresponding array / object sub-fields) accept http:// / https:// URLs, mediakit://... file_ids, or local file paths
• http:// / https:// URLs and mediakit://... file_ids are submitted as-is; local file paths are first uploaded by the CLI as mediakit://... file_ids and then submitted to the cloud tool
• Parameter descriptions in each tool's reference are sourced from the APIHub/OpenAPI raw field descriptions; even when they show public URLs or HTTP/HTTPS URLs, that only reflects the resource form the cloud API ultimately receives — it does not restrict the CLI's ability to pre-process local paths in cloud mode

Local-mode notes

• Local output directory priority: --output-path > MEDIAKIT_OUTPUT_PATH > config output_path > ~/.mediakit/temp
• When --output-path points to a concrete media filename, it is used as the final output file directly; otherwise the filename is built as {source_name}_{tool_name}.{ext}, with a 6-digit random suffix appended on collision
• When no filename can be extracted from the input URL or path, fall back to {tool_name}-{UnixNano}.{ext}
• Local mode depends on ffmpeg / ffprobe; when missing, the error response includes an install_guide
• Local-mode media processing output must conform to the corresponding interface response schema; internal execution metadata must not be emitted

Error responses

• CLI cloud mode forwards the original error object returned by the API as-is, without extracting message
• CLI local mode returns a structured error: {"error":{"type":"...","code":"...","message":"..."}}
• MCP error_response forwards the original error content as-is; a dict is used directly as the error field value

Idempotency parameter maintenance

Additional rules:

• client_token must not exceed 64 characters
• callback_args is useful for callback passthrough and reconciliation tracking

Polling strategy