Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
.claude/skills/creminiai-skill-creator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 38% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -14% | 0% |
Create new skills and iteratively improve them inside this SkillPack. Determine where the user is in the create → draft → test → improve cycle and help them move forward.
All skills live under {{SKILLS_PATH}}/<skill-name>/SKILL.md and config is at {{PACK_CONFIG_PATH}}. These paths override any generic advice. Never create skills in the current workspace directory — always use {{SKILLS_PATH}}.
Extract answers from the current conversation first, then fill gaps with targeted questions. Confirm before writing the first draft:
Clarify edge cases, input/output formats, success criteria, and dependencies as needed before writing test prompts.
Create the skill at {{SKILLS_PATH}}/<skill-name>/SKILL.md. Example template:
yaml--- name: example-skill description: "Analyze competitor pricing pages and generate a comparison matrix. Use when the user wants to benchmark pricing tiers, feature gaps, or positioning against specific competitors." --- # Example Skill ## Workflow 1. Collect the target URLs from the user. 2. Extract pricing tiers, features, and limits from each page. 3. Generate a comparison matrix as a markdown table. ## Output Return a markdown table with one column per competitor and one row per feature.
The description is the primary triggering mechanism — make it concrete with both what the skill does and when to use it. Keep the body focused on workflow, decisions, and output expectations. Place deterministic helpers under scripts/ and long reference material under references/.
Preserve the existing skill name when improving unless the user explicitly requests a rename.
After creating or updating a skill, sync {{PACK_CONFIG_PATH}}. Read the final SKILL.md, parse the YAML frontmatter, and upsert into the skills array:
json{ "name": "<frontmatter.name>", "description": "<frontmatter.description>", "source": "./skills/<frontmatter.name>" }
All three fields must come from frontmatter — do not guess from memory. Update existing entries instead of creating duplicates.
Prefer imperative instructions. Structure skills as: purpose, trigger guidance, required inputs, step-by-step workflow, output format, edge cases. For multi-domain skills, organize references by variant and tell the model how to choose.
After drafting the skill, propose 2–3 realistic test prompts phrased as a real user would.
If the user wants evaluation, run the prompts, compare outputs against expectations, note failures, and revise. Otherwise, do at least a lightweight sanity check before calling the skill complete.
When updating an existing skill:
name unless the user explicitly asks to rename itSKILL.md first{{PACK_CONFIG_PATH}}Focus on general improvements rather than overfitting to one example. Keep the prompt lean and remove instructions that are not earning their place.
Before you say the work is done, verify all of the following:
{{SKILLS_PATH}}/<skill-name>/SKILL.mdSKILL.md has name and description frontmatter{{PACK_CONFIG_PATH}} has a matching entry in skillssource field is ./skills/<skill-name>Other measured skills in the registry, with their headline benchmark lift.