agent-factory

I've been digging through skill implementations and came across your agent-factory—the recursive task delegation pattern is clever, though I'm curious how you're handling agent state synchronization at scale since that's where most systems start to fracture.

Links:

• GitHub
• Your agentic skill in the SkillzWave marketplace

The TL;DR

You're at 76/100, solidly in C territory—which is a good foundation but leaving meaningful points on the table. This is based on Anthropic's best practices for skill structure and usability. Your strongest area is Utility (16/20) where you're actually solving a real problem with structured approaches. The weak spots are Progressive Disclosure Architecture (20/30) and Writing Style (7/10), which are fixable.

What's Working Well

• Solid utility and problem-solving: You're addressing a genuine gap (custom agent creation) with structured workflows and concrete agent types (simple, agentic, supervisor). That's 16/20 points well-earned.
• Metadata and conventions: Name format is correct, YAML frontmatter is valid, and you maintain consistent terminology throughout (kebab-case, YAML structure, MCP integration patterns).
• Comprehensive examples: You've got extensive examples showing different agent configurations—that earned you bonus points for being grep-friendly and exemplary in scope.

The Big One: File Structure is Defeating Progressive Disclosure

Here's the problem: Your SKILL.md is 269 lines and your template (AGENTS_FACTORY_PROMPT.md) is 1,124 lines. That's massive. Progressive Disclosure means users should get the concept quickly from SKILL.md (~80 lines), then drill deeper into references. Right now, they're climbing a mountain to understand what your skill does.

The fix: Split your template into smaller, purpose-built files:

• templates/agent-prompt-core.md (core rules, ~200 lines)
• templates/examples/ (separate .md per example type)
• templates/reference-matrix.md (tool access matrix)
• Add a TOC to any file over 100 lines

Trim SKILL.md to overview only—move detailed workflows, agent type explanations, and MCP integration details to references/. This alone should gain you +4 points on PDA and improve usability.

Other Things Worth Fixing

1. Add trigger phrases to your description (currently missing): "Use when asked to 'create agent', 'build sub-agent', 'generate custom agent', or 'new agent'". Right now people have to know it's '@agent-factory' to find it. Should gain +2 points.

2. Voice consistency: You're mixing imperative and second-person ("This skill helps you", "you create"). Pick imperative throughout: "Creates custom agents" instead of "This skill helps you create". Small fix, +1 point.

3. Feedback loops are weak: You mention "validation" but don't show the run→test→fix cycle for generated agents. Add a concrete validation workflow: "Test generated agent with claude --agent {name} --dry-run, fix YAML errors, verify tool access, test invocation." +1 point.

4. Trim marketing language: Watch for promotional terms ("comprehensive system", "seamlessly integrate") that don't belong in technical documentation. Keep it instructional. Minor but affects writing score.

Quick Wins

• Most impactful: Restructure SKILL.md + template into modular files → +4-5 points
• Discoverability: Add trigger phrases to description → +2 points
• Polish: Fix voice inconsistencies + strengthen feedback loops → +2 points
• Result: Realistic path to 85-88 points with focused effort

Checkout your skill here: SkillzWave.ai | SpillWave We have an agentic skill installer that install skills in 14+ coding agent platforms. Check out this guide on how to improve your agentic skills.