Skills as Specialized Roles¶
Skills Encode Judgment¶
A skill is a set of reusable instructions that tells your AI coding assistant how to do a task your way. Not generically — your way. With your conventions, your quality standards, your edge cases handled.
The Blue Square track teaches skills through the "We Do, You Do" pattern: work through a task together with AI, refine the output until it matches what you'd produce, then capture those instructions as a reusable file. Test with a fresh context to prove the skill is self-contained. Refine based on gaps.
Here's what most practitioners miss: the skill file isn't the value — the refinement loop is. A skill that was captured after one iteration encodes surface-level process. A skill that was refined through five iterations encodes your judgment about edge cases, failure modes, and quality tradeoffs. The quality of the skill directly reflects how thoroughly you refined it.
From Personal Tools to Team Assets¶
In the Blue Square track, skills solve a personal problem: "I keep re-explaining the same convention." In the Black Diamond track, skills solve a team problem: role gaps.
Your team of four doesn't have a dedicated PM, a dedicated QA engineer, or a dedicated designer. But you still need those functions. Skills let you turn each role gap into an AI role:
- A decomposition skill that breaks epics into independently shippable stories with acceptance criteria — the PM role
- A test generation skill that translates Given/When/Then criteria into automated tests — the QA role
- A UI component skill that enforces your design system's conventions — the designer role
- A code review skill that checks for your team's specific anti-patterns — the tech lead role
Each skill encodes the judgment that role would bring. The AI coding assistant executes the process; the skill provides the expertise.
Skills live in .claude/skills/ as markdown files. Claude Code reads skill metadata automatically and loads full instructions when a task matches. Skills are now part of the open Agent Skills standard — the same skill file works across tools.
Skills are implemented as custom instruction files or prompt templates. Reference them when working on related tasks.
Skills are prompt templates invoked by name. Create them for repeatable processes and invoke with /name.
Version Control and Sharing¶
Skills are files. Files live in version control. That means:
- Every teammate gets the same skills — they travel with the project, not with the person who wrote them
- Skills evolve through pull requests — refinements are reviewed, discussed, and approved like any other code change
- Skills compose — a deployment skill can reference a test generation skill, building a pipeline from modular pieces
The shift from "personal tool" to "version-controlled team asset" is what separates Blue Square skills from Black Diamond skills. When a senior practitioner captures their judgment in a skill, every contributor to the project — human and AI — benefits from that judgment immediately.
Team Discussion: Role Gaps¶
Format: Team Discussion Time: ~2 minutes
Look at your team. You're building a multi-zone analysis and alerting engine. You need data ingestion, risk analysis, alert logic, and a dashboard — plus tests, deployment verification, and data validation.
Discuss: What roles are missing from your team? If you had a dedicated QA engineer, what would they check? If you had a data engineer, what conventions would they enforce? Which of those could be captured as a skill — and which ones require human judgment that a skill can't encode?
Key Insight¶
Skills are repositories of judgment made explicit through iteration. The "We Do, You Do" pattern isn't one-and-done — the value comes from continued refinement. At the Black Diamond level, skills solve a team-scale problem: turning role gaps into AI roles so that every contributor operates with the same quality standards. The skill file travels with the project, not with the person who wrote it.