Autonomous 4-Panel Manga Agent with ComfyUI

How to Use:
To start, open the package in your favorite AI agent interface (Claude Code, Cursor, Cline, etc.) and instruct the agent:
"Load the skill defined in SKILL.md and generate a 4-panel manga from this plot: [Your 1-line story plot]"

The agent will immediately:

1. Decompose your plot into a structured 4-panel narrative sequence (Introduction, Development, Twist, and Resolution).
2. Inject your character's visual anchor prompts (constants like facial features and outfit) and frame-specific action prompts (variables) into the API workflow JSON.
3. Invoke the bundled Python orchestrator to send the prompt to ComfyUI.
4. If ComfyUI rejects the run due to a missing checkpoint, the agent will query your active ComfyUI instance for available checkpoints, patch the JSON automatically, and retry.
5. Monitor the render progress and report the finished file paths when complete.

---
About Autonomous 4-Panel Manga Agent
The Autonomous 4-Panel Manga Agent is a production-grade AI agent skill package designed for creators and developers to automate multi-frame comic strip production. Instead of manually editing ComfyUI workflow graphs or dealing with fragile terminal scripts, this package equips your AI agent to operate as a headless director for your local ComfyUI API.

Conventional image automation projects suffer from environment mismatch errors, fragile folder-watching loops, and character drift across panels. This skill overcomes all three by teaching the agent to directly query ComfyUI’s internal configuration schema, poll the API task history safely using client-side unique IDs, and decompose prompt parameters using consistent anchor variables.

The package includes the core SKILL.md rules, a detailed setup guide, a self-repairing Python client script, and an optimized 4-panel workflow JSON configured for multi-frame generation.

Key Highlights:

• Self-Healing Error Correction: If the generation is rejected due to a checkpoint mismatch, the agent automatically queries the live ComfyUI instance (via the /object_info endpoint), finds a valid checkpoint model installed on your machine, patches the file, and retries.
• Safe Task Tracking: Uses prompt unique IDs (prompt_id) and the ComfyUI history polling API instead of unstable directory watchers, preventing collisions and race conditions.
• Visual Consistency Anchors: Decomposes prompts into constants (face, outfit, hair style, LoRAs) and variables (poses, emotions, action descriptions) to maintain high character fidelity across panels.
• Real-time Progress Indicator: The client script outputs terminal progress dots (. . . .) while ComfyUI is rendering, giving both you and your agent visual feedback on the task status.