One Prompt Lets GenericAgent Turn 31 WeChat Groups’ Daily Chat into a Single Poster

When asked who prepares their daily community report, the answer is GenericAgent (GA) – an autonomous AI that completes the entire workflow from a single natural‑language instruction.

1. Find the SOP

GA first reads its own operation manual wechat_group_daily_report_sop.md to understand the reporting format, then searches for examples of the requested "old‑school American tattoo flash" style to know the visual appearance.

2. Extract messages from 31 groups

GA reads the local WeChat database, filters out irrelevant chats, and locks onto 31 business groups. It exports the day’s messages, totaling 79 entries, then runs a Python script to clean the data (remove garbled text, count speakers, rank the most active groups, and list top contributors).

3. Two‑step summarization

Using a large model, GA performs a two‑step summarization on the 79 fragmented messages, extracting five hot discussion topics, three key highlights, and six keywords, producing a structure that looks as if a human spent half an hour organizing it.

4. Reverse‑engineer the image API

The poster‑generation step requires the gpt-image-2 endpoint, which lacks public documentation. GA downloads the front‑end JavaScript of the service, uses regular expressions to locate hidden parameters such as task_id and images/generations, and extracts the call pattern. It then iteratively tests different endpoints and parameter combinations until the API returns a correct response.

5. Assemble the report and generate the poster

Before rendering the image, GA compiles a structured daily report that includes a data overview, today’s highlights, hot topics, and keywords. The poster is then generated in the requested tattoo‑flash aesthetic—brown‑paper texture, classic colors, modular layout—using the previously reverse‑engineered API. The final poster displays the same text that GA wrote.

The generated report shows, for example, 79 total messages from 20 speakers, the most active groups, three key highlights (goal‑loop skill on SkillHub, mobile GA CLI support, timed news collection), five discussion topics, and six keywords (GA, mobile, Feishu, CLI, Galley, model).

Continuous operation

GA produces such posters daily; past editions are shown in a series of images. The system demonstrates that an autonomous agent can read real data, write code, debug tools, and deliver a finished product without human hands, and it can be repurposed for any repetitive, multi‑tool workflow.

In short, GA is not merely a chatbot; it is an autonomous agent that can ingest real data, generate code, troubleshoot unknown problems, and deliver a complete artifact, making it suitable for any repetitive, cross‑tool task.