skill-creator

Skill Creator

A skill is a modular, self-contained package of capabilities and instructions that extends the agent's abilities. Skills follow the standard Agent Skills specification and reside in /skills/<skill_id>/ (persisted in ~/.ollama-agent/skills/<skill_id>/).

Directory Structure

A skill directory must contain a SKILL.md file and may include optional subdirectories:

/skills/<skill-id>/
├── SKILL.md                 # Required: Frontmatter YAML + instructions
├── scripts/                 # Optional: Executable Python or Bash helper scripts
├── references/              # Optional: Deep reference documentation, schemas, API specs
└── examples/                # Optional: Input/output samples, templates, test cases

Critical Decision: When to Create scripts/ vs Pure SKILL.md

When designing a skill, always evaluate whether helper scripts are needed:

1. Create Helper Scripts (scripts/) when:

   • The task involves deterministic algorithms, parsing structured formats (JSON, AST, XML, CSV), regex transformations, or data validation.
   • The task performs repetitive Git workflows, CLI tool wrapping, API fetching, or complex file operations.
   • The operation is error-prone or wasteful to do purely through LLM token generation.
   • Language: Write modular, self-contained Python scripts (or Bash if simple shell piping is sufficient). Ensure scripts handle errors gracefully and output clear diagnostic messages.

2. Use Pure Markdown (SKILL.md) when:

   • The skill focuses on reasoning guidelines, architectural reviews, code style heuristics, tone/formatting advice, or general workflow orchestration without deterministic code execution.

SKILL.md Specification

1. Frontmatter (YAML)

Every SKILL.md must start with YAML frontmatter:

---
name: skill-id-or-name
description: A clear, concise 1-3 sentence summary of what this skill does and the exact trigger conditions for when the agent should activate it.
---

IMPORTANT (Level 1 Discovery): The description is loaded into the agent's discovery prompt before the full skill content is read. It MUST explicitly state what the skill does and when the agent should load and follow it.