Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Skill Creator

An end-to-end workflow for drafting new Claude skills, testing them with evals and benchmarks, and tuning their descriptions for reliable triggering.

Dev & CodingAdvanced★ 578⑂ 72AI score 9/10Last updated: Sep 26, 2026

What it does

  • Drafts the skill: interviews you about intent, triggers, and output format, then writes SKILL.md frontmatter and body following progressive-disclosure conventions.
  • Builds and runs test cases: creates 2–3 realistic prompts, saves them to evals/evals.json, and spawns with-skill plus baseline (no-skill or old-version) runs in parallel.
  • Quantitative benchmarking: grades assertions, aggregates pass rate, duration, and token usage (mean ± stddev) via scripts.aggregate_benchmark, and launches a browser review viewer (eval-viewer/generate_review.py) with Outputs and Benchmark tabs.
  • Iteration loop: reads feedback.json, generalizes from the feedback instead of overfitting, and reruns into a new iteration directory.
  • Description optimization: generates 20 should-trigger / should-not-trigger queries, runs scripts.run_loop with train/held-out split, and applies the best description.
  • Packaging: optionally builds a .skill file via scripts.package_skill.

Who it's for

  • Developers and automation folks authoring their own Claude Code skills.
  • Anyone whose skill under-triggers and needs a sharper description.
  • Teams that want prompt quality measured with numbers, not vibes.

Example uses

  1. "Turn this Excel monthly-revenue workflow into a skill" → interview → draft → 3 test runs with baselines → review in the viewer → revise.
  2. Improving an existing research-helper skill: snapshot the old version and A/B it against the edit to prove the change helps.
  3. A skill that never fires: build tricky near-miss negatives plus varied positives and run the description optimization loop.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/scenario-labs/skills/HEAD/.agents/skills/skill-creator/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the .agents/skills/skill-creator folder from the GitHub repo scenario-labs/skills into my ~/.claude/skills/scenario-labs-skill-creator/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/scenario-labs/skills.git /tmp/sl-skills && mkdir -p ~/.claude/skills && cp -r /tmp/sl-skills/.agents/skills/skill-creator ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal (you need git and Python 3).
  2. Clone the repo: git clone https://github.com/scenario-labs/skills.git /tmp/sl-skills
  3. Create the skills folder: mkdir -p ~/.claude/skills
  4. Copy the skill: cp -r /tmp/sl-skills/.agents/skills/skill-creator ~/.claude/skills/
  5. Verify ~/.claude/skills/skill-creator/SKILL.md exists along with the scripts/, eval-viewer/, agents/, and references/ folders.
  6. Restart Claude Code and say something like "I want to create a new skill" or "optimize this skill's description" to trigger it.
  7. Note: the eval/benchmark path assumes Claude Code with subagents and permission to run Python; on Claude.ai only the reduced workflow described in the skill applies.
View source on GitHub ↗License: MIT