skills

HYDRA SKILLS

What This Folder Is

This is where Hydra's knowledge lives — every domain, every proven approach, every pattern it knows. Drop a TOML file and Hydra learns a new domain. No training. No fine-tuning. No code.

Rules

1. One folder per skill — named clearly (e.g., architecture/, finance/, security/)
2. genome.toml is required — contains [[entries]] with situation + approach pairs
3. functor.toml is optional — maps domain concepts to axiom primitives (Risk, Understanding, Dependency, Volume)
4. Hydra reads on boot — skills are loaded into the genome store at startup
5. You can add skills at any time — restart Hydra to load new ones
6. Hydra writes here too — the self-writing genome creates entries in skills/generated/

Structure

skills/
  README.md              ← this file
  architecture/
    genome.toml          ← 10 proven approaches
    functor.toml         ← axiom mappings
  finance/
    genome.toml          ← 26 proven approaches
    functor.toml
  developer/
    genome.toml          ← 30 proven approaches
    functor.toml
  ... (29 skills total, 303 genome entries)

Format

# genome.toml
[[entries]]
situation    = "choosing between microservices and monolith"
approach     = "start with a monolith — extract services only when boundaries are clear"
confidence   = 0.91
observations = 5000

See SKILL-FORMAT.md in this folder for the complete specification.

How It Works

When you ask Hydra a question, the genome store queries all loaded skills using IDF-weighted scoring + axiom vector cosine similarity. The best matches are injected into the prompt as proven approaches — with mathematically grounded confidence intervals.

Drop a TOML file. Hydra learns a new domain. That is it.