Preface¶

When an Agent receives tasks such as “search,” “open this site,” or “scrape a webpage,” it tends to directly invoke web_search, web_fetch, browser, or opencli. However, if the search API is not configured, the interface returns an error, or the site requires login, the common approach is to repeatedly retry the same tool or launch a browser prematurely. This wastes steps and easily leads to missing still-available structured scraping paths.

sfe (SkillHub coordinates @user_9ca58e78/ddd) introduced below is a procedural Skill designed for AI Agents: it requires reading its instructions before invoking the aforementioned network-related tools and mandates executing error handling and degradation sequences as agreed upon therein. It originates from the SkillHub community (publisher user_9ca58e78), is categorized under ai-agent, is currently version 1.0.0, and includes the tag requires_api_key=true.

What is this¶

sfe is not a crawler script for a specific website, but rather a set of operational constraints and process instructions written into SKILL.md. The official description positions it as a protocol that must be followed before calling web_search, web_fetch, browser, or opencli; failing to read this Skill can lead to mishandling failure scenarios and overlooking available tools.

SkillHub Page: https://www.skillhub.cn/skills/user_9ca58e78/ddd
API Entry (Data Side): https://api.skillhub.cn/user_9ca58e78/ddd

Core Features¶

The following capabilities are derived from the introduction published on SkillHub. After installation, the full text of SKILL.md within the Skill package should be treated as the standard.

Mandatory Reading Before Invocation¶

Before an Agent prepares to use any capability of web_search, web_fetch, browser, or opencli, it should first load and follow the steps in sfe, rather than directly calling tools based on default habits.

Error Handling on Search Failure¶

When web_search fails, the protocol requires guiding the user to complete API-related configuration, rather than failing silently or retrying infinitely. This aligns with the Skill tag requires_api_key=true: web search often relies on external APIs, and if unconfigured, the issue should be explicitly handed back to the user.

Fallback Chain¶

The introduction defines a structured fallback path: before utilizing the browser, prioritize structured fallback scraping via the 70+ sites covered by the opencli CLI. This implies that when the primary path (such as search or general fetch) is unavailable, one should first attempt the set of sites handled by opencli, and then consider the browser, to reduce heavy and slow browser automation.

Site Login Entries¶

The Skill contains login URLs organized by site, facilitating the handling of authentication by the Agent or user via the correct entry point in scenarios requiring login, instead of blindly opening the homepage.

Trigger Intent¶

When the following intents appear in the user or task, sfe should be associated and loaded, for example: search, go online, look up information, open website, scrape webpage, news, hot topics, as well as English expressions like web search / fetch / browser / opencli, etc.

Installation and Enablement¶

The official SkillHub installation instructions are available at https://skillhub.cn/install/skillhub.md. The core point is: when installing skills, you must use --dir to point to the skills directory actually used by the current Agent; otherwise, it may install to the default ./skills/ and not be recognized.

1. Check or Install SkillHub CLI¶

First, confirm whether skillhub exists on the local machine:

command -v skillhub && skillhub --version

If not installed, you can follow the official documentation to perform a complete installation (CLI + default Skills):

curl -fsSL https://skillhub-1388575217.cos.ap-guangzhou.myqcloud.com/install/install.sh | bash

If only the CLI is needed:

curl -fsSL https://skillhub-1388575217.cos.ap-guangzhou.myqcloud.com/install/install.sh | bash -s -- --cli-only

2. Install sfe to the Agent’s skills directory¶

Install the coordinate @user_9ca58e78/ddd into the directory corresponding to the current environment. Common path examples given in the SkillHub documentation:

• Cursor: ~/.cursor/skills/
• Claude Code: ~/.claude/skills/
• Codex: ~/.codex/skills/ or .agents/skills/ under the project

The installation command takes the following form (please replace <skills directory> with the path used by your Agent):

skillhub install @user_9ca58e78/ddd --dir <skills directory>

Alternatively, follow the prompts on the SkillHub skill page to let the AI complete the installation based on skillhub.md, with the target coordinate also being @user_9ca58e78/ddd.

3. Enablement and API¶

After installation, the Skill needs to be enabled within the Agent (the specific entry point varies by client; refer to the Skills configuration of the product you are using). Since the tag is requires_api_key=true, before using capabilities like web_search, you should complete the configuration for search-related APIs according to sfe and the product documentation; in case of failure, guide the user to complete the configuration as required by the Skill.

Typical Usage¶

The official page does not provide separate business code examples; the reproducible workflow comes from the introduction and the Skill type itself. It is recommended to implement it in the Agent session in the following order.

1. The user raises a network-related requirement (e.g., “check a news hot topic” or “open a site and scrape the body text”).
2. The Agent reads sfe (SKILL.md) first, then selects tools, rather than directly using web_search / browser.
3. Prioritize trying the corresponding primary path in the introduction (e.g., web_search or web_fetch); if web_search fails, follow the protocol to prompt the user to configure the API, do not pretend it succeeded.
4. If the primary path is still insufficient, follow the fallback chain to try opencli (structured capabilities covering 70+ sites), and then consider using browser depending on the situation.
5. If the target site requires login, refer to the site login URLs within the Skill to handle the login entry according to the protocol.

Users can explicitly trigger it, for example: “Help me search for a certain topic and scrape the official page according to sfe protocol,” so that the Agent loads this Skill before calling tools.

Use Cases and Notes¶

Who is it for

• Frontline developers who frequently use Agents like Cursor, Claude Code, and Codex for searching, scraping pages, and tracking hot topics.
• Teams wishing to unify “how to explain search failures to users” and “when to use opencli vs. when to launch a browser,” using a single Skill to constrain Agent behavior.

Limitations and Notes

• The details of the protocol are subject to the SKILL.md within the installed package; this article only summarizes the public introduction from SkillHub and does not elaborate on every login URL or opencli subcommand within the site.
• SkillHub Statistics (Community Source): Rating 100000, Downloads 1539, Installs 0 (based on scraped data, page display may be updated).
• Dependency on web_search requires self-preparation and configuration of the API; sfe emphasizes that failures must guide configuration, rather than skipping it.
• Incorrect installation directories will prevent the Agent from reading the Skill; be sure to use skillhub install ... --dir <skills directory>.

Conclusion¶

The value of sfe lies in writing “how to call network tools, what to do upon failure, and what to degrade to” into the mandatory reading protocol for the Agent, reducing issues such as misuse of web_search, premature browser automation, and ignoring opencli structured fallbacks. For installation or updates, refer to the SkillHub skill page and installation documentation: https://www.skillhub.cn/skills/user_9ca58e78/ddd.