SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillslaohan-douyinsousuo

laohan-douyinsousuo

Data & analytics

View repository

Updated 10-15 22:12

About this skill

Douyin Search

This is the Douyin search orchestration interface in the workflow. It only uses the host's already installed and already logged-in OpenCLI, and does not include its own crawler, browser, cookie copying, independent profile, or Python runtime.

Input

  • query: required, the search keyword.
  • limit: optional, 1-30, default 10.
  • episode: optional, this episode's episodes/<slug> path; only writing to 00-抖音搜索证据.json and .mdis allowed, and writing to candidates, source-health, or 00-选题.*。

Execution

1. Precheck

opencli doctor
opencli douyin whoami -f json

Execute the search when Browser Bridge is connectable and logged_in=true . If any precheck fails, in episode mode still write out UNAVAILABLE evidence and the original error for ① to continue judging; for an independent search request, truthfully report that it is currently unavailable. Do not extract or copy Chrome cookies, and do not install new tools.

2. Search

In episode mode, use the deterministic wrapper to execute 1-3 shortlist queries at a time and atomically write JSON/Markdown:

node ~/Documents/laohan-skills/laohan-douyinsousuo/scripts/search-evidence.mjs \
  --episode episodes/<slug> \
  --query "关键词一" \
  --query "关键词二" \
  --limit 10

An independent single query can be executed directly:

opencli douyin search "$QUERY" --limit "$LIMIT" -f json

The result must be a JSON array, and must preserve item by item the rank、desc、author、url、aweme_id, publish time, and available interaction fields returned by OpenCLI. For stable id, prioritize using aweme_id or the /video/<id>in the URL; only when neither is present use sha256(url + desc)[:12]. Ranking must not participate in the stable id; do not reorder by missing fields, and do not describe OpenCLI's result order as "ranked by likes".

3. Failure degradation

Execute in order, and stop as soon as one succeeds:

  1. Add --trace retain-on-failureto the same command, and use OpenCLI autofix to repair the existing adapter; a maximum of 3 attempts in total.
  2. If opencli doctor is still normal but the adapter continues to fail, use opencli browser douyin-search bind to bind to the user's currently logged-in Douyin tab, and only perform reading and evidence collection. Before executing the browser degradation, explain that it will operate on the current tab.
  3. If it still fails, record the failure layer and the original error. In episode mode, mark the platform data UNAVAILABLE and return the evidence file, and do not escalate it to a blocker for ①; for an independent search request, report that no results were obtained this time. Do not temporarily install DrissionPage, Playwright, a browser, or a self-made crawler.

Output and judgment

First show the search health status and the actual number of returned items, then show the result table:

# Author Likes Title Link
1 author likes/unavailable desc url
  • If an interaction field is missing, write null and in field_availability mark false; if the actual returned number is 0 , then keep 0 and mark true, and do not conflate the two.
  • When an empty array is returned, write EMPTY_OR_FIELD_UNAVAILABLE; this cannot be used to assert that there is no related content on Douyin.
  • The currently installed OpenCLI 1.8.6's douyin search only provides search result samples, and cards usually only have the likes field;douyin hashtag hot can provide site-wide hot topics, and is already used by laohan-xuanti's discovery route.douyin hashtag search currently fails in actual testing for the interface,hashtag hot --keyword Nor can they be reliably filtered by keyword. Therefore, search volume, total supply, growth rate, and official content gap all default to UNAVAILABLE; they are used only when OpenCLI actually returns these fields in the future, and are not inferred from the number of results or likes.
  • Factual conclusions must be verified against the original video or authoritative sources; search results can only serve as discovery evidence.
  • When the user separately requests search analysis, the visible content types and recurring angles in the results may be summarized; it must not determine the final topic of the workflow, and must not fabricate facts not provided by the search results.
  • When specifying episode , the runtime-locked wrapper must be used: first write the machine source of truth 00-抖音搜索证据.json, then generate .mdfrom it; Markdown must not add conclusions that the JSON does not contain.

Minimal Episode JSON contract:

{
  "schema_version": 1,
  "collected_at": "ISO-8601",
  "executor": {
    "name": "opencli",
    "version": "实际版本",
    "doctor_status": "PASS|FAILED",
    "logged_in": true,
    "availability_status": "AVAILABLE|UNAVAILABLE"
  },
  "queries": [{
    "query": "关键词",
    "command": "opencli douyin search ... -f json",
    "attempted_at": "ISO-8601",
    "status": "OK|EMPTY_OR_FIELD_UNAVAILABLE|FAILED",
    "result_count": 1,
    "results": [{"id":"稳定id","rank":1,"desc":"原字段","author":"原字段","url":"https://..."}]
  }]
}

OK must result_count > 0 and equal the length of results;EMPTY_OR_FIELD_UNAVAILABLE must be 0 and an empty array; FAILED must be 0, an empty array, and have error. When all queries FAILED or precheck fails, write availability_status: UNAVAILABLE, and still complete episode evidence persistence; ① do not infer from this that the platform has no demand.

Boundaries

  • Downloading videos, blogger work lists: use /laohan-xiazai。
  • Comments, publishing, account operations: use /laohan-yunying or the corresponding OpenCLI adapter.
  • DrissionPage,browser_cookie3 cookie injection, independent user-data-dir, additional Chrome/Chromium, and background automatic publishing or replies are prohibited.