Skill Quality Improvement

Problem Description

A team maintains an image-processor skill that helps agents work with their internal image transformation pipeline. A recent skill review flagged several issues: the description lacks trigger conditions, some code examples use outdated flag syntax, and a few sections repeat guidance from the reference documentation verbatim.

Your task is to improve the skill based on the review feedback below. The skill bundle includes a REFERENCE.md file with detailed API documentation — you may read it for context, but your goal is to improve the skill's quality and conciseness. Produce the improved skill file as output.

Output Specification

Produce:

• SKILL.md — the improved version of the skill file
• change_log.md — a summary of changes made and why each improves the skill

Input Files

The following files are provided as inputs. Extract them before beginning.

=============== FILE: skill_bundle/SKILL.md ===============

name: image-processor description: | Processes and transforms images using our internal pipeline. Supports resizing, format conversion, watermarking, and batch operations.

Image Processor

Transform images using the imgctl command-line tool.

Resize Images

To resize an image, the system needs to know the target dimensions. The width and height should be specified in pixels. Aspect ratio can be preserved or ignored depending on the use case.

imgctl resize --input photo.jpg --width 800 --height 600 --output resized.jpg

Available resize modes (from REFERENCE.md):

• fit — scale to fit within dimensions, preserving aspect ratio
• fill — scale and crop to fill exact dimensions
• stretch — stretch to exact dimensions, ignoring aspect ratio

Convert Format

To convert between formats:

imgctl convert --input photo.jpg --format webp --quality 85 --output photo.webp

Supported formats: JPEG, PNG, WebP, AVIF, GIF, TIFF (from REFERENCE.md)

Quality parameter: 1-100, higher is better quality but larger file. Default: 85.

WebP typically achieves 25-35% smaller file sizes than JPEG at equivalent quality.

Batch Processing

Process multiple files at once:

imgctl batch --input-dir ./photos --output-dir ./processed --operation resize --width 1200

The batch command reads all image files from the input directory. Supported input file extensions: .jpg, .jpeg, .png, .webp, .gif, .tiff, .avif (from REFERENCE.md).

Error Handling

imgctl exits with code 0 on success, 1 on error. Error details are written to stderr.

Common errors:

• INPUT_NOT_FOUND — input file or directory doesn't exist
• UNSUPPORTED_FORMAT — input format not supported
• PERMISSION_DENIED — can't write to output path =============== END FILE ===============

=============== FILE: skill_bundle/REFERENCE.md ===============

Image Processor — Complete Reference

Resize Modes

• fit — scale to fit within dimensions, preserving aspect ratio
• fill — scale and crop to fill exact dimensions
• stretch — stretch to exact dimensions, ignoring aspect ratio

Supported Formats

Input and output: JPEG, PNG, WebP, AVIF, GIF, TIFF

Format Conversion

Quality range: 1-100 (default 85). WebP achieves ~25-35% smaller files than JPEG.

Batch Processing

All image files in input dir are processed. Supported extensions: .jpg, .jpeg, .png, .webp, .gif, .tiff, .avif

Full Flag Reference

imgctl resize

imgctl convert

imgctl batch

Exit Codes

• 0: Success
• 1: Error (details in stderr)

Error Codes

• INPUT_NOT_FOUND
• UNSUPPORTED_FORMAT
• PERMISSION_DENIED
• WRITE_FAILED =============== END FILE ===============

=============== FILE: review_feedback.txt =============== === Skill Review: image-processor === Overall Score: 64%

ISSUES: [ERROR] Description missing "Use when" trigger clause [WARNING] Several sections inline content already in REFERENCE.md

Dimension Scores: Completeness: 1/3 (33%) - Missing "Use when" trigger Actionability: 3/3 (100%) - Good executable examples Conciseness: 2/3 (66%) - Inline reference duplication reduces score Robustness: 2/3 (66%) - Error codes present but no retry patterns =============== END FILE ===============