---
title: Skill Alchemist
category: product
entity_type: agent
price: $79
canonical: https://forgehouse.ai/agents/skill-alchemist/
lang: en
hreflang_alt: https://forgehouse.ai/tr/ajanlar/skill-alchemist/
last_updated: 2026-06-20
---

# Skill Alchemist

> Skill upgrader to Ultra (Pro+) standard

An agent that upgrades existing skills to a higher standard. It measures against a rubric, adds edge cases, anti-patterns, cross-skill awareness and mental models, and promotes recurring patterns into durable rules. It scores every skill against a quality rubric before handing it back, and rejects generic mental-model padding outright.

## Use cases
- Upgrading a thin skill to production quality
- Adding edge cases and anti-patterns to a skill
- Injecting cross-skill awareness and mental models
- Auditing a set of skills against a quality rubric
- Promoting repeated patterns into rules

## Benefits
- Skills that handle the hard cases, not just the happy path
- Consistent quality across your whole skill library
- Knowledge captured once, reused forever
- A measurable bar instead of subjective polish

## What’s included
- Skill upgrades to the 7-dimension Ultra standard with skill-specific content
- Cross-skill linking: minimum two typed connections per skill, bidirectional
- Mental-model sections that fit THIS skill: generic SOLID/DRY stubs rejected
- Edge-case, anti-pattern and exception-handling additions to existing skills
- Monthly pattern-promotion: recurring lessons become permanent skill content
- Quality rubric scoring before any skill ships back

## Who it’s for
Teams maintaining a skill library who want a consistent quality bar.

## How it runs
Upgrading a skill is a three-phase loop: score it against a 12-criterion rubric, transform without deleting working content, then re-score until the threshold is genuinely met. Every claim is proven with command output, never impression.
1. Reads the three mandatory reference files before writing anything: the canonical skill template skeleton, the cross-skill linking protocol and the enforcement rules that ban template reuse and stub mental models. It only ever touches existing skill files; creating new skills belongs to a sibling tool.
2. Phase one, resonance: reads the full skill and its references, counts the live skill ecosystem with a real directory listing instead of quoting a stale number, scans for overlapping skills, then scores the skill against a 12-criterion rubric covering persona depth, tactical detail, anti-patterns, agency integration, the 7-dimension matrix and cross-skill links.
3. Phase two, alchemy: transforms by fixed principles. Working content is never deleted, only extended; generic personas become specific experts; checklist items get the what, why, how and when-not-to; every tactic gains at least one cross-dimension note and one applied mental model; the file stays under 500 lines with overflow moved to reference files.
4. Phase three, synchronization: re-reads the transformed skill and re-scores the rubric. Below 11 of 12 means back to phase two; the loop repeats until the threshold is genuinely met, not declared.
5. Proves every claim with command output: line counts, code block counts and link counts come from real word counts and pattern searches, never from impression. An upgrade claim without that evidence is invalid.
6. Reports before-and-after rubric scores, what was added and how the skill's description changed for routing, then hands the enriched skill to the security auditor for a malicious-pattern scan. One skill at a time, never several in parallel.

## FAQ
### My skills are thin one-pagers, is that a workable starting point?
Yes, that's the core use case. It takes an existing thin skill, measures it against the quality rubric, then adds edge cases, anti-patterns, cross-skill links and skill-specific mental models until it clears the Ultra bar.

### How does it decide a skill is actually done?
Every skill gets scored against the rubric before it ships back. Generic mental-model padding: a bare SOLID or DRY stub: is rejected outright, and each skill needs at least two typed, bidirectional cross-skill connections. The bar is measurable, not subjective polish.

### Does it create new skills from scratch?
No. Its scope is upgrading skills that already exist; creation is a separate job. What it adds beyond depth is a monthly pattern-promotion pass, where recurring lessons get turned into permanent skill content instead of being relearned.

## Price
$79, one-time, no subscription. VAT included.

Related guide: [AI and LLM engineering](https://forgehouse.ai/guides/ai-llm-engineering/)