Agent Skill
2/7/2026

inference-operator-and-batch-yaml

Bridge skill between creative prompting and batch inference pipeline. Use when converting refined prompts into batchable YAML jobs for the LoRA inference harness. Covers YAML job specs, structured outputs, title conventions, and integration with comfy_automation.

D
davidrd123
2GitHub Stars
1Views
npx skills add davidrd123/ComfyPromptByAPI

SKILL.md

Nameinference-operator-and-batch-yaml
DescriptionBridge skill between creative prompting and batch inference pipeline. Use when converting refined prompts into batchable YAML jobs for the LoRA inference harness. Covers YAML job specs, structured outputs, title conventions, and integration with comfy_automation.

name: inference-operator-and-batch-yaml description: > Bridge skill between creative prompting and batch inference pipeline. Use when converting refined prompts into batchable YAML jobs for the LoRA inference harness. Covers YAML job specs, structured outputs, title conventions, and integration with comfy_automation. allowed-tools: Read, Grep, Glob, Write

Inference Operator & Batch YAML

A bridge skill for converting creative prompts into structured YAML batch jobs for the inference pipeline.

YAML Job Specification

Reference: 99-yaml-format.txt

Basic Structure

jobs:
  - title: "Descriptive title reflecting subtext"
    prompt: |
      Full prompt text using agreed architecture...

  - title: "Another variation title"
    prompt: |
      Variation prompt text...

Title Conventions

  • Titles should reflect subtext, not just describe the action
  • Use evocative language that captures the feeling/intent
  • Keep titles concise but meaningful

Prompt Architecture

Prompts should follow the established architecture from prompt-engineering-toolkit:

  1. Front-load critical elements (subject, action, setting)
  2. Use appropriate stylization/lexicon
  3. Include aesthetic control keywords
  4. End with thematic capstone if needed

Frequently Used Prompts

Reference: 03-inference-prompts.md

This document contains practical examples of:

  • Structured inference prompts
  • Multiple variants with approach variations
  • Title and prompt pairing patterns

Output Workflow

  1. Refine prompts using prompt-engineering-toolkit
  2. Structure as YAML job entries
  3. Write to batch file for pipeline consumption

Integration Points

This skill outputs YAML that can be consumed by:

  • comfy_automation.py batch runner
  • Gemini Batch API (per ADR-033)
  • Other harnesses that consume the same YAML spec

Example Output

jobs:
  - title: "The Weight of Silence"
    prompt: |
      Graffito Mixed-Media Stop-Motion — A close-up of a figure's
      hands, trembling slightly as they grip the edge of a worn table.
      The lighting is low, a single overhead lamp casting harsh shadows.
      Materials: painted cardboard, wire armatures, fabric scraps.
      Palette: muted browns and grays with a single warm highlight.
      The scene breathes with quiet tension and unspoken words.

  - title: "The Weight of Silence (alt: softer)"
    prompt: |
      Graffito Mixed-Media Stop-Motion — A close-up of weathered hands
      resting gently on a sun-warmed table surface. Soft diffused light
      from a nearby window. Materials: sculpted paper pulp, cloth textures.
      Palette: warm cream tones with gentle shadows.
      A moment of peaceful resignation.
Skills Info
Original Name:inference-operator-and-batch-yamlAuthor:davidrd123