How I Used Agent Plan to Automate the Final Step of Writing a WeChat Public Account Post

Writing a WeChat public account article often stalls at the cover‑image stage because finding a suitable, copyright‑free picture is time‑consuming.

Agent Plan, Volcano Engine's first multi‑modal subscription package, fills this gap by bundling text, code, image, and video models (including GLM‑5.1, Kimi‑K2.6, DeepSeek) together with a real‑time search quota and an embedding‑vision model, allowing a single subscription to handle diverse AI tasks.

To use it, the author configures Hermes Agent to point to the Agent Plan endpoint. The configuration steps are:

Select Custom endpoint and enter the URL https://ark.cn-beijing.volces.com/api/plan/v3.

Provide the dedicated Agent Plan API KEY.

Choose the text‑generation model doubao-seed-2.0-lite as the model name.

Set the display name to volcengine-agent-plan.

After saving the settings, the author installs the official image‑generation skill by downloading the ZIP from

https://findskill.com/volcengine/agentplan/byted-ark-seedream-skill

and extracting it to ~/.hermes/skills/.

With the skill in place, the workflow becomes a single prompt. For example, sending “Help me make a public‑account cover, the article is about Agent Plan, title is ‘Agent Plan’” returns four candidate images with the title rendered; omitting the title yields a pure background. The author can also supply a reference image to guide the style.

On first execution the system asks for the API KEY; after that the generated images appear on the desktop. The tool can produce multiple candidates at once, and the user may replace an unsatisfactory result with a custom reference image.

Beyond cover images, Agent Plan supports generating videos, writing code, and other multimodal tasks, demonstrating how a unified subscription that aggregates multiple models can dramatically expand what users can accomplish.