agentsclimarketplace

Creating skills

Skill rerelurelu/skillbox/skills/creating-skills

Install
npx -y skills add rerelurelu/skillbox --skill creating-skills

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Authors new Agent Skills following the agentskills.io specification, covering SKILL.md frontmatter requirements, naming conventions, progressive disclosure patterns, and validation. Use when the user asks to create, author, write, add, design, or scaffold a new skill, or to convert an existing slash command into a skill.

The file declares its own license as GPL-3.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.5 KB, as published. Nobody here has run it

Creating Agent Skills

Author new Agent Skills following the agentskills.io specification.

Quick Reference

Directory structure:

skills/<skill-name>/
├── SKILL.md              # Required entry point
└── *.md                  # Optional reference files (one level deep only)

Required frontmatter:

---
name: <gerund-form-name>          # lowercase, hyphens, max 64 chars
description: "<third-person>..."  # what + when, max 1024 chars, with trigger keywords
---

Recommended frontmatter:

license: "GPL-3.0"
user-invocable: true              # also expose as slash command
allowed-tools: "Read Edit Bash"   # space-delimited STRING, not array

Process

Step 1: Read the full guide

Read AUTHORING.md (bundled, sourced from Anthropic's official skill authoring docs). It covers progressive disclosure, anti-patterns, evaluation, and examples in depth.

Step 2: Choose a name

Use gerund form (verb + -ing) for discoverability and consistency:

  • Good: processing-pdfs, analyzing-spreadsheets, creating-skills
  • Avoid: vague names (helper, utils), reserved words (anthropic, claude)

Step 3: Write a discovery-friendly description

The description is what the agent uses to decide whether to load the skill. Be specific and include trigger keywords.

Bad: description: "Helps with code"

Good: description: "Removes AI-generated slop (unnecessary comments, defensive checks, style inconsistencies) from uncommitted changes. Use after AI-assisted edits, before committing, or when the user asks to deslop or clean up generated noise."

Always write in third person. The description is injected into the system prompt; first/second person causes discovery problems.

Step 4: Author the body

  • Keep SKILL.md body under 500 lines. Split longer content into sibling files.
  • Match degrees of freedom to task fragility (specific scripts for fragile ops, high-level guidance for flexible ones).
  • Reference files must be one level deep from SKILL.md (avoid A → B → C chains).

Step 5: Validate

gh skill publish --dry-run

Must report zero errors. Common errors:

  • allowed-tools must be a string — convert YAML array to space-delimited string
  • recommended field missing: license — add license: "GPL-3.0"
  • name/description validation — check 64/1024 char limits and allowed character sets

Step 6: Update repo metadata

  • Add the skill to README.md's skill table
  • Add the skill to the install instructions section

Step 7: (Optional) Publish

gh skill publish --tag vX.Y.Z

This creates a GitHub release with provenance metadata, enabling pinned installs and update tracking.

Common Pitfalls

PitfallSymptomFix
allowed-tools as YAML listgh skill publish --dry-run errorsConvert to space-delimited string
First-person descriptionSkill doesn't auto-trigger reliablyRewrite in third person
Description lacks triggersSkill never auto-loadsAdd "Use when..." with concrete keywords
SKILL.md over 500 linesSlow loading, context bloatSplit into reference files
Nested references (one file links to another that links to another)Agent reads incompletely (head -100)Flatten so all refs are one level from SKILL.md
Time-sensitive infoBecomes wrong over timeMove to "old patterns" section or remove
Windows-style paths (scripts\helper.py)Breaks on UnixUse forward slashes always

Anti-patterns

  • Adding "voodoo constants" without justification
  • Punting errors to the agent rather than handling them in scripts
  • Offering many alternative approaches when one default suffices
  • Mixing terminology (field / box / element for the same concept)

See AUTHORING.md §Anti-patterns for the full list.

Reference

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.