Batch Skill Creation

Overview

Template-driven batch creation of Hermes skills using Python scripts that generate SKILL.md files directly. Produces 500-700 skills per run with built-in validation, deduplication, and description formatting.

When to Use

• "Bulk skill creation for large catalog expansion"
• "When creating 50+ skills efficiently"
• "Template-driven skill generation at scale"
• "Automated skill creation with validation"

Key Approaches

1. Set Up the Environment

import os
SKILLS_DIR = "C:/Users/dubem/AppData/Local/hermes/skills"
existing = set(d for d in os.listdir(SKILLS_DIR) if os.path.isdir(os.path.join(SKILLS_DIR, d)))

2. Define the Skill Factory Function

def create_skill(name, desc, tags, related, overview):
    # Validate description: ≤60 chars, starts "Use when", ends with "."
    assert len(desc) <= 59, f"DESC TOO LONG: {name} ({len(desc)} chars)"
    assert desc.startswith("Use when"), f"Bad desc start: {name}"
    assert desc.endswith("."), f"Bad desc end: {name}"
    
    related_str = ', '.join([f"'{r}'" for r in related.split(',') if r]) if related else "general"
    # ... build body and markdown

3. Domain-Based Template Pattern

Define domains as tuples (prefix, label, overview, tags), then generate 4-8 variations per domain:

domains = [
    ("supply-chain", "Supply Chain", "Manage supply chains.", "supply-chain, logistics, procurement"),
    ("logistics", "Logistics", "Plan logistics ops.", "logistics, delivery, shipping"),
    # ... 80+ domain templates
]

for prefix, label, overview, tags in domains:
    for suffix, desc_suffix in [
        ("-fundamentals", f"for {label.lower()} fundamentals."),
        ("-implementation", f"for {label.lower()} implementation."),
        ("-optimization", f"for {label.lower()} optimization."),
        ("-management", f"for {label.lower()} management."),
    ]:
        name = prefix + suffix
        desc = "Use when " + desc_suffix
        if len(desc) > 59: desc = desc[:56] + "."
        # Create skill...

4. Deduplication Check

if name in existing:
    continue  # Skip existing skills

5. Standard SKILL.md Template

Every generated skill follows the standard format:

• YAML frontmatter: name, description, version, author, license, platforms, metadata (tags, related_skills)
• ## Overview — brief domain overview
• ## When to Use — bullet-point trigger conditions
• ## Key Approaches — numbered implementation steps
• ## Common Pitfalls — numbered list of common mistakes
• ## Verification Checklist — markdown task list

6. Validation Before Creation

def desc_ok(d):
    return len(d) <= 59 and d.startswith("Use when") and d.endswith(".")

7. Run and Verify (PROVEN PATTERN: execute_code over subagents)

# Use execute_code for direct execution - NO subagent delegation
# Subagents calling skill_manage per-skill cause HTTP 524 timeouts
from hermes_tools import execute_code
# ... or write batch script and run via terminal
# python batch_script.py

8. Template-Driven Expansion (4 SKILLS PER DOMAIN)

suffixes = [
    ("-fundamentals", "for {label.lower()} fundamentals."),
    ("-implementation", "for {label.lower()} implementation."),
    ("-best-practices", "for {label.lower()} best practices."),
    ("-troubleshooting", "for {label.lower()} troubleshooting."),
]

for prefix, label, overview, tags in domains:
    for suffix, desc_suffix in suffixes:
        name = prefix + suffix
        desc = "Use when applying " + desc_suffix.format(label=label)
        if len(desc) > 59: desc = desc[:56] + "."
        make_skill(name, desc, tags, overview)

Common Pitfalls

1. Description exceeds 60 chars — Always truncate to desc[:56] + "." and validate with len(desc) <= 59
2. Missing 'Use when' prefix — Descriptions must start with "Use when" exactly
3. Missing period at end — All descriptions must end with "."
4. Duplicate skill names — Always check existing set before creating
5. Script too large for write_file — Use execute_code to write large Python scripts to disk, or split into smaller batch scripts
6. Subagent timeout for bulk creation — Subagents that call skill_manage per-skill (one API call per skill) cause HTTP 524 timeouts at 120s proxy read timeout or 1200s hard ceiling. For 50-100+ skills, delegation always fails. Always use direct terminal python batch_script.py execution instead.
7. Variable scope errors in retry scripts — When patching batch scripts, ensure counter variables (created, failed) are initialized before any loops that reference them
8. F-string escaping in Python — When embedding the SKILL.md template, be careful with curly braces in YAML/metadata sections
9. Not cleaning up scripts — Always remove temporary batch scripts after the session

Reference Files

• references/template_batch_creator.py — Reusable starter script; edit the DOMAINS list and run
• scripts/verify_skills.py — Post-creation verification: checks YAML, sections, description length Produces 4 variations per domain with automatic validation.

Verification Checklist

• Script validates descriptions before creation
• desc_ok() function checks length, prefix, and period
• existing set prevents duplicate creation
• Each skill has ## Overview, ## When to Use, ## Key Approaches
• Each skill has ## Common Pitfalls and ## Verification Checklist
• Total SKILL.md count matches expected number
• Temporary batch scripts cleaned up after use
• All generated descriptions are ≤60 characters