dongchedi-work-search

Dongchedi content keyword search

Search Dongchedi content by keyword, outputting title, author, plays, likes, comments, publish time, and full content link.

Trigger scenarios

Use this Skill when the user raises the following requests:

"Dongchedi keyword search"
"Search Dongchedi content/videos/image-text posts"
"Check content about a certain car series on Dongchedi"
"What content does a certain brand have on Dongchedi"
"Keywords: Zunjie / Xiaomi / Zhijie"

If the search returns no results, only output "No data available" and possible reasons; do not proactively replace the keyword to query other content.

Keyword suggestions and normalization

Keyword length suggestions

Input keywords are recommended to be no more than 6 characters, with 2 characters being optimal (such as 零跑、智界); overly long combined words easily trigger interface timeouts.
When the user provides a long combined word, guide them to split it into short entity words: first query the brand/car series word (such as 零跑), then query the model word as needed (such as T03）。
Inputting multiple keywords at once separated by spaces, commas, etc. is not supported (such as 零跑智界、零跑,智界); when encountering such input, prompt the user that this Skill queries only one keyword at a time, ask the user to choose one of them before executing, and do not automatically concatenate multiple keywords into one query.

Keyword normalization (triggered only when the query fails)

The interface responds unstably to long combined words (such as 零跑 T03 车道保持走不了直线), while short words are the most stable. Only when the query fails (timeout/network error/task submission failure), retry with normalization in the following order:

Retry after removing spaces: remove all spaces from the keyword, such as 零跑 T03 → 零跑T03。
Retry after reducing to a single entity word: if it still fails, reduce it to one entity word, such as 零跑T03 → 零跑(prioritize keeping the brand/car series word; pure English or numeric model names such as T03 may keep the model word).
If both normalization retries fail: output the failure explanation and the list of keywords already tried, and do not continue retrying.

Constraints:

A normal interface return (including empty workList) does not count as a failure and does not trigger normalization; empty results are handled as "No data available", and other keywords are not queried instead.
The normalization retry chain is at most 2 times; apart from the two methods "remove spaces" and "reduce to a single entity word", it must not be replaced with other unrelated keywords.
Normalization applies only to long combined words describing the same object (such as brand + model + issue description); multiple independent keywords separated by spaces/commas (such as 零跑智界) are not eligible for normalization, and the user should be prompted to choose one keyword at a time before executing.

Data description

Output content

Content list: output search results as a Markdown table, including serial number, title, author, content ID, plays, likes, comments, publish time, and full content link
Content type: supports two content types, "Comprehensive" and "Video", with Comprehensive as the default
Work Link: Use a complete clickable link to directly access the work
Fixed Reminder: Include a reminder about obtaining the API Key in the output

Sorting and Criteria

Output in descending order by play count by default; when specified by the user, it can be changed to likes, comments, time, or the original order of the data service
Publish time is automatically converted to local time for display; if conversion fails, display it as is
Only display the fields returned by the search results, and do not fabricate body text, cover images, or detail information

One-Click Installation

Prerequisites

Python 3.8+ (only the standard library is used, no additional dependencies need to be installed)
Red Fox Data API Key (go to Red Fox hub to register and obtain it)

Installation Steps

Place the skill folder into your Skills directory
Configure the environment variable REDFOX_API_KEY(for configuration details, see references/core_workflow.md）
No additional dependencies need to be installed; use it directly

Workflow

For the core execution process (intent recognition, script invocation, table rendering rules, empty results and error handling, etc.), see references/core_workflow.md. The Agent must follow all rules in that file when executing. For the complete field descriptions, see references/api-spec.md。

Usage Guide

Directly describe the query requirements in natural language, no need to memorize commands.

Intent	Example Wording	Effect
Keyword Search	"Help me search for works related to Zhijie on Dongchedi"	Output a list of works by keyword (by default in descending order by play count)
Specify Video	"Search Dongchedi for Xiaomi SU7 videos"	Only output works of the video type
Specify Count	"Check Dongchedi Chengdu Auto Show content, show the first 10 items"	Only display the first 10 items
Specify Sorting	"Sort by number of likes and take a look"	Change to output in descending order by likes
Continue Paging	"Next page"	Output the next 20 items

Notes

Data Authenticity: All values come from the data service returns and are displayed truthfully; estimation or fabrication is strictly prohibited
Do Not Replace When the Query Yields No Results: When the result is empty, only output "No data available" and the possible reasons, and do not proactively switch to querying other keywords or other works; only when the query fails may normalization retries be performed according to the "Keyword Suggestions and Normalization" section
Credit Protection: Each successful query may consume credits; unless explicitly specified by the user, do not proactively page through or batch query multiple keywords
Authentication: For first-time use, you need to configure the environment variable REDFOX_API_KEY. For obtaining and configuring the API Key, see references/core_workflow.md
Link Integrity: The work links in the output should remain complete and clickable, and URLs truncated by the terminal should not be used