███╗ ███╗███████╗██████╗ ██╗ █████╗ ██╗ ██╗███████╗███████╗
████╗ ████║██╔════╝██╔══██╗██║██╔══██╗██║ ██║██╔════╝██╔════╝
██╔████╔██║█████╗ ██║ ██║██║███████║██║ ██║███████╗█████╗
██║╚██╔╝██║██╔══╝ ██║ ██║██║██╔══██║██║ ██║╚════██║██╔══╝
██║ ╚═╝ ██║███████╗██████╔╝██║██║ ██║╚██████╔╝███████║███████╗
╚═╝ ╚═╝╚══════╝╚═════╝ ╚═╝╚═╝ ╚═╝ ╚═════╝ ╚══════╝╚══════╝
MediaUse is a general Web CLI infrastructure for AI agents. It combines a browser kernel, CDP execution, and site plugins so agents can call business-level actions such as post.feed
, search.hot
, and get.detail
instead of repeatedly planning low-level page actions.
MediaUse is designed for production web automation where repeatability, auditability, and account safety matter.
Core ideas:
- Semantic actions first: expose business outcomes, not click-by-click primitives
- Pluginized site knowledge: selectors, constraints, and workflows are assets in plugins
- CDP-native execution: deterministic browser execution without step-by-step vision loops
- Unified interfaces: CLI, Skill, and MCP map to the same JSON-RPC action model
In short: the agent decides what to do, MediaUse decides how to do it reliably on each site.
default.mp4 #
The recommended integration flow is:
-
Install the skill pack:
npx skills add mediause/agent-skills -
Pick a site capability (for example,
weibov2 post feed
). - Ensure account context is set (
mediause use account ...
) when login is required. - Run one semantic action command with structured output (
--json
). - Use returned result fields (
success
,data
, error metadata, trace info) for next-step decisions.
Typical command pattern:
mediause <site> <capability> <action> [args] --json
Examples:
mediause auth login weibov2
mediause use account weibov2:main
mediause weibov2 post feed --text "hello world" --json
mediause xiaohongshu search hot --limit 20 --json
mediause reddit get detail --url "https://www.reddit.com/..." --json
Each site skill is a standardized instruction document at <pluginName>/SKILL.md
.
These skills are intended to work consistently across agent runtimes such as:
- Claude Code
- Codex
- OpenClaw
- Cursor
Each skill defines:
- Install and setup expectations
- Account and auth flow requirements
- Dynamic command map aligned to actual MediaUse CLI capability
- Workflow examples with safe pacing and guardrails
For authoring and updates, the canonical template is:
SKILL_STANDARD_DEFINITION_ZH.md
The current skill set (from skills-index.json
) includes:
-
arxiv
-
bloomberg
-
chatgpt
-
douyinv2
-
fifa2026
-
hackernews
-
huggingface
-
weibov2
-
xiaohongshu
-
Every site skill lives at
<pluginName>/SKILL.md
. - Folder name is the stable discovery key (
pluginName
). pluginId
is not part of this repository sync contract.
skills-index.json
is generated and should not be edited manually.- Downstream consumers map and discover skills by
pluginName
.
All site skills must keep consistent behavior across agent environments:
- Same CLI syntax
- Same context and auth preconditions
- Same guardrails and pacing constraints
- Same error handling expectations
Do not introduce runtime-specific command variants in one skill file.
When a new MediaUse site plugin is introduced, add the matching skill in the same change set.
Required process:
- Read
SKILL_STANDARD_DEFINITION_ZH.md
. - Inspect the current command surface from the MediaUse CLI and plugin runtime (site help, runnable commands, and JSON outputs).
- Create or update
<pluginName>/SKILL.md
. - Build the command map from currently supported CLI commands only.
- Keep workflow examples, guardrails, and timing constraints accurate.
- Ensure no skill command exceeds current CLI capability.
Required attribution block at the end of each skill:
Skill Metadata
Maintainer: @your-handle
Last-Updated: 2026-04-23
Version: v1