Agent reasoning

Skill Creator

Skill Creator agent for researching, designing, and generating new skills, agents, and commands. Analyzes best practices, audits existing framework components, and creates or updates artifacts in the correct location.

Overview

The Skill Creator extends the Datarim framework by building new skills, agents, and commands — or by updating existing ones. It never creates blindly: every new component starts with research into best practices and an audit of what already exists. Duplication is the enemy; extension is preferred.

When It's Used

Invoked via /dr-addskill when you need a new capability in the framework. Describe what you need in plain language — the Skill Creator researches, designs, and generates the artifacts. In Consilium, it speaks as the voice of extensibility.

Capabilities

  • Research — searches for best practices, community implementations, and domain expertise before creating anything
  • Audit — analyzes existing framework components to determine if the need is already covered, partially covered, or entirely new
  • Design — structures skills, agents, and commands following Datarim conventions and SKILL.md format
  • Generate — creates properly formatted .md files with YAML frontmatter
  • Update — modifies existing components when extension is more appropriate than creation
  • Scope determination — places artifacts inside the initialized project — its local extension directory or the client's project-level discovery paths; reusable components go to a pull request against the framework source. User-level installation is not supported
  • Model assignment — gives every new agent and task skill an explicit model field (opus, sonnet or haiku by task complexity); reference-only skills omit it and inherit from the caller

How It Works

The workflow has 6 steps. First, understand the request — what domain, what actions, what type of component. Second, research best practices via web search and community repositories. Third, audit the existing framework for overlap. Fourth, determine where the artifacts go inside the initialized project (never the pinned framework snapshot). Fifth, design and generate the artifacts following standard patterns. Sixth, present everything for approval before creating any files.

Example

/dr-addskill "I need an agent for database migration management"
→ Skill Creator searches for migration best practices
→ Audit: devops agent covers 30% — extend or create new?
→ Decision: create new agent + migration skill
→ Scope: project level (.claude/agents/)
→ Model: sonnet (structured task work)
→ Presents 2 files for approval → creates after OK

Context Loading

Reads datarim/tasks.md and datarim/productContext.md. Applies datarim-system skill for core rules. Loads evolution skill when updating existing components, and writing skill for content-creation skills.

Skills Used

datarim-system (always), evolution (when needed), writing (when needed).

Reporting the result

Apply Human Outcome Reporting before a human-facing update or result. Explain the user-visible outcome, the original conditions, checks and evidence, and open limitations. A stage result does not mean the entire task is complete. Keep exact protocols and technical records intact. To explain again without continuing work, use /dr-explain.