gemini

I took a look at your gemini skill and wanted to share some thoughts.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 79/100, which lands you in C territory — solid fundamentals with real utility, but there's low-hanging fruit to push this into B range. Based on Anthropic's best practices, your strongest area is Spec Compliance (13/15) — the frontmatter and metadata are clean. The weak spot is Progressive Disclosure Architecture (17/30) — you're loading all 250 lines every time someone invokes this, regardless of what they need.

What's Working Well

• Trigger phrases are excellent — "pipe content to Gemini", "second opinion", "parallel analysis", "headless AI task" — that's 6 specific hooks that make this discoverable. Search will find it.
• The core pattern is crystal clear — cat input | gemini "PROMPT" --model gemini-2.5-pro > output.md. Simple, memorable, copy-pasteable.
• You've thought about model selection — splitting gemini-2.5-pro (complex reasoning) from gemini-2.5-flash (speed) is the right mental model. Users will appreciate having that choice explained upfront.
• Spec compliance is tight — Valid YAML, correct naming conventions, solid description. No metadata debt here.

The Big One: Token Bloat from Monolithic Structure

Your skill is 250 lines in a single SKILL.md file. That means every time Claude invokes this skill, they load:

• Overview
• Quick Start
• Core patterns
• Model selection
• Advanced examples
• Error handling
• Troubleshooting

...even if they just need the basic "pipe to Gemini" pattern.

The fix: Split this into a layered architecture:

• Keep SKILL.md at ~100 lines with: metadata, overview, quick-start pattern, when-to-use, and one simple example
• Create references/advanced-patterns.md for the workflow patterns and deeper examples
• Create references/model-selection.md for gemini-2.5-pro vs. gemini-2.5-flash deep dive

This could gain you +6 points on PDA alone. It's the difference between "I load everything" and "I load only what you need."

Other Things Worth Fixing

1. The --model flag gets repeated 15+ times — Establish once in Core Command Pattern: "All examples use gemini-2.5-pro unless noted. For speed, substitute gemini-2.5-flash." Then drop it from most examples. Saves tokens, improves readability. +2 points

2. Use cases show commands but no numbered workflows — "Alternative Perspective" example just shows a command. Make it: (1) Save Claude's analysis, (2) Run gemini command, (3) Verify output, (4) Compare both analyses. Explicit steps beat implicit patterns. +2 points

3. No complete workflow template — Add a fill-in-the-blank section: INPUT_FILE=, PROMPT=, OUTPUT_FILE=, then the command. Users can copy, modify one line, go. +1 point

Quick Wins

• Restructure for PDA — Split into SKILL.md + references/ (biggest bang for buck, +6 points)
• Consolidate model flags — State once, omit from examples (+2 points)
• Add numbered step workflows — Turn examples into procedures (+2 points)
• Create template section — One copy-paste-modify workflow (+1 point)

Hit these and you're looking at 90/100 territory.

Checkout your skill here: SkillzWave.ai | SpillWave We have an agentic skill installer that install skills in 14+ coding agent platforms. Check out this guide on how to improve your agentic skills.