Skip to content
Back to skills

Template

ASecurity

One or two sentences that say WHAT this skill does AND WHEN the agent should load it. Write in the third person, name concrete tasks, and include trigger keywords a user would actually say. Example: "Generate professional release notes from a git history. Use when cutting a release, drafting a changelog, or summarizing commits since the last tag."

  • 45 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 6, 2026
ai-agentsgogit

Security analysis

A100/100

Scanned October 6, 2026

npx -y skills add frankxai/claude-skills-library --skill template --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Template?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Template
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/frankxai-template-claude-skills-library/badge)](https://www.skillsdirectory.com/skills/frankxai-template-claude-skills-library)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: your-skill-name
description: One or two sentences that say WHAT this skill does AND WHEN the agent should load it. Write in the third person, name concrete tasks, and include trigger keywords a user would actually say. Example: "Generate professional release notes from a git history. Use when cutting a release, drafting a changelog, or summarizing commits since the last tag."
---

# Your Skill Name

> Delete this template's comments as you fill it in. Keep the body under ~500 lines —
> push deep material into `references/`, deterministic helpers into `scripts/`, and
> templates/fonts/icons into `assets/`. See `../spec/README.md` for the full standard.

## When to use this

Describe the situations this skill is for, in one tight paragraph. Be specific — this
reinforces the `description` and helps the model trigger correctly.

## Instructions

Give the agent clear, opinionated guidance: the workflow, the decisions, the gotchas.
Prefer numbered steps for fragile or order-dependent procedures, prose for open ones.
Assume the reader is a capable engineer — don't explain the basics.

1. Step one.
2. Step two.
3. Step three.

## References (optional)

Link to deeper docs the agent should read only when needed:

- `references/deep-dive.md` — loaded on demand (link it once the file exists).

## Scripts (optional)

Describe any executable helpers and whether the agent should **run** or **read** them:

- `scripts/do_thing.py` — run to perform X deterministically.

## Examples (optional)

Show one or two worked examples so the agent can pattern-match the desired output.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.