skill-creator

name	skill-creator
description	Authors and structures professional-grade agent skills following the agentskills.io spec. Use when creating new skill directories, drafting procedural instructions, or optimizing metadata for discoverability. Don't use for general documentation, non-agentic library code, or README files.

Skill Authoring Procedure

Follow these steps to generate a skill that adheres to the agentskills.io specification and progressive disclosure principles.

Step 1: Initialize and Validate Metadata

1. Define a unique name: 1-64 characters, lowercase, numbers, and single hyphens only. This must exactly match the parent directory name (e.g., name angular-testing must live in angular-testing/SKILL.md).
2. Draft a description: Max 1,024 characters, written in the third person. Include positive triggers ("Use when...") and negative triggers ("Don't use for...").
3. Execute Validation Script: Run the validation script to ensure compliance before proceeding: python3 scripts/validate-metadata.py --name "[name]" --description "[description]"
4. If the script returns an error, self-correct the metadata based on the stderr output and re-run until successful.

Step 2: Structure the Directory

1. Create the root directory using the validated name.
2. Initialize the following subdirectories:
   • scripts/: For tiny CLI tools and deterministic logic.
   • references/: For flat (one-level deep) context like schemas or API docs.
   • assets/: For output templates, JSON schemas, or static files.
3. Ensure no human-centric files (such as README.md, CHANGELOG.md, INSTALLATION.md) are created.

Step 3: Draft Core Logic (SKILL.md)

1. Use the template in assets/SKILL.template.md as the starting point.
2. Write all instructions in the third-person imperative (e.g., "Extract the text," "Run the build").
3. Apply Specific, Procedural Formatting:
   • Use Step-by-Step Numbering: Define the workflow as a strict chronological sequence. Clearly map out decision trees (e.g., "Step 2: If source maps are required, run scripts/build.sh. Otherwise, skip to Step 3.").
   • Provide Concrete Templates: If an output has a specific structure, place a template in assets/ and command the agent to copy its structure instead of describing it in text.
   • Use Consistent, Specific Terminology: Use identical, domain-native terms throughout (e.g., in Angular, use "template" consistently instead of "html" or "view").
4. Enforce Progressive Disclosure:
   • Keep the main SKILL.md under 500 lines.
   • If a procedure requires a large schema, complex rule set, or bulky context, move it to a flat file in references/.
   • Command the agent to read the specific file only when needed (JiT loading): "Read references/api-spec.md to identify the correct endpoint."

Step 4: Identify and Bundle Scripts

1. Identify "fragile" tasks (regex, complex parsing, or repetitive boilerplate).
2. Outline a single-purpose script for the scripts/ directory designed as a tiny CLI with arguments.
3. Ensure the script uses standard output (stdout/stderr) to communicate success or failure. Write descriptive, human-readable error messages so the agent knows exactly how to self-correct without user intervention.

Step 5: Skill Composition (Optional)

If a skill is complex, compose it using router/subskill patterns:

1. Create a high-level router skill.
2. Define conditional triggers or procedures referencing other specific skill folders (e.g., "To build the client, see [path to client-build skill]. To build the server, see [path to server-build skill].").

Step 6: Final Logic Validation

1. Review the SKILL.md for "hallucination gaps" (points where the agent is forced to guess).
2. Verify all file paths are relative and use forward slashes (/).
3. Cross-reference the final output against references/checklist.md.

Error Handling

• Metadata Failure: If scripts/validate-metadata.py fails, identify the specific error (e.g., "STYLE ERROR") and rewrite the field to remove first/second person pronouns.
• Context Bloat: If the draft exceeds 500 lines, extract the largest procedural block and move it to a file in references/.