Hierarchical Skills Taxonomy for Agentic Systems

[Zochory]

Why This System Exists

Agentic systems fail when they cannot reliably compose, reuse, and audit the behavior they generate. This repo treats “skills” as first‑class, versioned artifacts on disk, with a consistent taxonomy and validation rules. The result is a system where capabilities are:

• discoverable (taxonomy paths)
• composable (dependencies + capabilities)
• auditable (metadata + SKILL.md + history)
• reproducible (workflow outputs can be cached)

Core Idea

Instead of a monolithic prompt or ad‑hoc tool chain, the system stores skills as structured folders and metadata inside a taxonomy. A DSPy‑based workflow then creates new skills from tasks and registers them in the taxonomy with validation and traceability.

Primary Use Cases

• Skill bootstrapping: generate a new skill from a user task and store it in the taxonomy.
• Capability standardization: enforce consistent metadata, structure, and documentation across skills.
• Skill discovery: search by taxonomy path and dependencies to assemble working sets.
• Operational reliability: run skill creation with caching and validation to avoid regressions.

Benefits

• Structured knowledge: skills are stable artifacts, not ephemeral prompts.
• Repeatable creation: the workflow follows a 6‑step process with validations.
• Controlled growth: taxonomy conventions keep the system scalable.
• Operational visibility: metadata + validation reports + cache stats aid debugging.
• Composable behavior: skills can declare dependencies and capabilities.

System Context

flowchart LR
  User[User/Operator] --> TUI[TUI / CLI]
  TUI --> Workflow[DSPy Skill Workflow]
  Workflow --> Taxonomy[(Skills Taxonomy on Disk)]
  Workflow --> LLM[LLM Provider]

Loading

[linear]

QRE-266