/dr-addskill

Overview

/dr-addskill extends the Datarim framework by creating new skills, agents, or commands — or updating existing ones. The Skill Creator agent researches best practices, audits the current framework for overlap, determines the correct scope (project-local extension or a framework pull request), and generates properly formatted artifacts.

Usage

/dr-addskill "Docker deployment best practices"
/dr-addskill "Interior design specification skill"
/dr-addskill "Code review agent for Python projects"

What It Does

1. Parse Request — understands what domain or capability is needed and whether it should be a skill, agent, command, or combination.
2. Research — searches for best practices in the target domain, examines existing Claude Code skill patterns from community repositories, and downloads 2-3 relevant examples for reference.
3. Audit Existing Framework — lists all current skills, agents, and commands. Checks if the need is already covered fully or partially. Decides: create new, update existing, or extend and supplement.
4. Determine Scope — uses the explicitly initialized project: local extensions go under .datarim-runtime/local/, and skills are exposed through the project's native client discovery directories. Reusable framework changes go to the source repository as a pull request. Global (user-level) installation is not supported.
5. Design — creates artifacts following Datarim conventions:
   • Skills: YAML frontmatter + structured markdown sections, under 500 lines
   • Agents: frontmatter with required model field + role statement + capabilities
   • Commands: frontmatter + instructions + next steps
6. Present for Approval — shows what files will be created or updated, why this structure was chosen, and how to invoke the new capability. Waits for approval before writing anything.
7. Test the Skill — for discipline-enforcing skills, a RED → GREEN → REFACTOR pass: observe the failure without the skill, write the smallest skill that answers the captured rationalizations, then re-run until no new loophole appears.
8. Apply and Confirm — creates directories and writes files. Reports what was installed, how to use it, and whether a reload is needed.

Arguments

A description of the needed capability. Required.

Quality Criteria

• Description is a trigger, not a summary — written for the model: "when should I fire?"
• Does not state the obvious — focuses on what pushes Claude out of its default behavior
• Does not railroad — gives goals and constraints, not prescriptive step-by-step
• Includes a Gotchas section for highest-signal model guidance
• Under 500 lines per skill file

Example Session

> /dr-addskill "Database migration safety checklist"

Researching: database migration best practices...
  Found: 3 relevant community skill examples
  Analyzed: zero-downtime migration patterns

Audit results:
  Existing skills: 79
  Related: compliance.md (partial overlap)
  Decision: Create new skill, reference from compliance.md

Scope: project-local (.datarim-runtime/local/)

Proposed files:
  1. .datarim-runtime/local/skills/db-migration/SKILL.md (new, 120 lines)
  2. .datarim-runtime/local/skills/compliance/SKILL.md (update: add reference)

Install? (y/n)

Related Commands

• /dr-optimize — audit framework after adding skills
• /dr-help — verify new command appears in the list

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.