Skill Editor

Create or edit skills based on user requirements.

Understanding Skills

Skills follow the AgentSkills.io open standard. Each skill is a directory containing a SKILL.md file with frontmatter and markdown instructions.

Key Features

• Directory-based: Each skill lives in .cursor/skills/<skill-name>/
• SKILL.md Required: Main instructions file with YAML frontmatter
• Project or Personal Scope: Project skills in .cursor/skills/, personal in ~/.cursor/skills/
• Optional Supporting Files: scripts/, references/, assets/ subdirectories

Directory Structure

.cursor/skills/<skill-name>/
├── SKILL.md           # Required - main instructions
├── scripts/           # Optional - executable code
├── references/        # Optional - additional docs
└── assets/            # Optional - templates, data files

Steps

Step 0: Determine Operation Mode

Analyze the request to determine whether this is a create or edit operation:

1. Edit Mode indicators:

   • File path mentioned (e.g., .cursor/skills/my-skill/SKILL.md)
   • Skill name with edit intent (e.g., "edit the deploy skill", "modify my-skill")
   • Explicit edit request with existing skill

2. Create Mode indicators:

   • Capability description (e.g., "skill for parsing JSON")
   • "create skill for...", "make a skill that..."
   • No existing skill matches

3. Ambiguous: Search .cursor/skills/**/SKILL.md, ask user to clarify

EDIT MODE

Step 1: Find the Skill

Search .cursor/skills/**/SKILL.md:

• If multiple matches, list and ask user to select
• If none found, offer to create a new skill

Step 2: Read and Analyze

• Load the SKILL.md file
• Identify frontmatter fields and content sections
• Understand current functionality

Step 3: Get Edit Type

Ask user:

• "What would you like to modify?"
• Options: "Update frontmatter" | "Modify instructions" | "Add supporting files" | "Change invocation settings"

Step 4: Apply Changes

• Apply modifications to the file
• Preserve sections not being changed
• Confirm modifications were successful
• Provide updated file path

Then STOP.

CREATE MODE

Step 1: Check for Duplicates

Search for similar skills:

• Search .cursor/skills/**/SKILL.md
• Search for similar functionality
• If found, ask: edit existing, create anyway, or cancel

Step 2: Generate Skill Name

Based on the request, suggest a name following these rules:

• Max 64 characters
• Lowercase letters, numbers, and hyphens only (a-z, 0-9, -)
• Must NOT start or end with hyphen
• No consecutive hyphens (--)
• Pattern: verb-noun or verb-noun-qualifier (e.g., parse-json, validate-email)

Confirm with user or let them provide custom name. Validate name before proceeding.

Step 3: Gather Requirements

Ask user:

Invocation - "Who should invoke this skill?" Options:

• "Both user and agent (Recommended)" - Default behavior
• "User only" - Add disable-model-invocation: true

Execution - "How should this skill run?" Options:

• "Inline (Recommended)" - Runs in current context

Step 4: Create Skill Directory and File

bash
1mkdir -p .cursor/skills/<skill-name>

Write .cursor/skills/<skill-name>/SKILL.md:

yaml
1---
2name: <skill-name>
3description: <What this skill does and when to use it. Include keywords that help identify relevant tasks. Max 1024 chars.>
4# Optional fields:
5# disable-model-invocation: true
6# license: MIT
7# compatibility: Requires git CLI
8# metadata:
9#   author: "Team"
10#   version: "1.0"
11---
12
13<Markdown instructions to follow when skill is invoked>
14
15## Steps
16
171. [First step - be specific about tools/methods]
182. [Second step]
193. [Validation/error handling]
20
21## Examples
22
23### Example 1: [Scenario]
24**Input:** `[input]`
25**Expected behavior:** [what should happen]
26
27## Notes
28
29- [Edge cases, limitations, best practices]

Step 5: Confirm Creation

Report:

• Skill name and location
• Invocation method (/skill-name or automatic)
• How to test it
• Mention editing with skill-editor for future changes

Then STOP.

SKILL.md Format Reference

Required Frontmatter

Optional Frontmatter

Quality Standards

Good skills:

• Focused (one task well)
• Clear description with trigger keywords
• Well-documented with examples
• Under 500 lines (move details to references/)

Name validation checklist:

• Lowercase only
• No spaces (use hyphens)
• No consecutive hyphens
• Doesn't start/end with hyphen
• Matches directory name
• Max 64 characters

Error Handling

Limitations (Cursor-specific)

Note: The following Claude Code features are NOT supported in Cursor:

• $ARGUMENTS string substitutions
• !`command` dynamic context injection
• argument-hint field
• user-invocable field
• allowed-tools field
• model field
• context: fork field
• agent field
• hooks field