hermes - 💡(How to fix) Fix [Bug]: baoyu infographic skill should not silently bypass image_generate [1 pull requests]

Bug Description

The bundled baoyu-infographic skill says to generate an infographic, but Step 6 can be interpreted loosely enough that an agent may substitute a deterministic handmade/scripted graphic instead of prompting the configured Hermes image_generate tool.

That is surprising for users because Hermes image generation may be explicitly configured, for example image_gen.provider=openai-codex and image_gen.model=gpt-image-2-high, and the user expects the infographic skill to exercise that backend.

Steps to Reproduce

1. Configure Hermes image generation, for example image_gen.provider=openai-codex and image_gen.model=gpt-image-2-high.
2. Ask the agent to create an infographic using the bundled baoyu-infographic skill.
3. Observe that the skill instructions say to generate an image, but do not explicitly forbid silently replacing the image-generation step with a scripted/handmade infographic when the agent tries to improve text readability.

Expected Behavior

The skill should clearly instruct the agent to use the Hermes image_generate tool with the assembled prompt from Step 5, so the configured image-generation backend is actually used.

If exact readable text is more important than image-model aesthetics, the agent should explain the trade-off to the user before switching to an HTML/SVG/Pillow or other deterministic overlay workflow.

Actual Behavior

The instruction currently says only:

Use the image_generate tool with the assembled prompt from Step 5.

That is directionally correct, but not explicit enough about avoiding silent substitution. In practice, an agent can bypass the configured image backend and create a scripted graphic while still claiming it created an infographic.

Affected Component

• Skills
• Image generation tool usage guidance

Proposed Fix

Tighten Step 6 in skills/creative/baoyu-infographic/SKILL.md and the bundled docs mirror to state that:

• the Hermes image_generate tool must be used with the assembled prompt;
• the skill is designed to prompt the configured Hermes image-generation backend;
• a scripted/handmade substitute must not be used silently;
• if exact readable text needs a deterministic overlay, the agent should tell the user before switching away from image_generate.

Duplicate Search

Searched existing issues and PRs for:

• baoyu-infographic image_generate
• infographic image generation skill
• baoyu infographic scripted
• infographic skill image_generate
• Hermes infographic gpt-image-2
• image_generate infographic skill

No matching open or closed issue/PR was found.

Debug Report

Not applicable. This is a bundled skill-instruction clarity bug, not a runtime crash.

Environment

• OS: Linux
• Hermes image generation config observed locally: openai-codex / gpt-image-2-high