Skills

Skills and agents — structured prompts that keep AI aligned with what was agreed on. No plugins, no extensions.

When you run a skill, Claude executes it with full context of your codebase and blueprint. Agent-powered skills go further, spawning parallel research agents to explore the codebase before acting. Clone the repo and everything is ready to use.

Product Owner

Commands to create, refine, and validate tickets using the contract template.

/draft

Create a new ticket using the contract template. Claude structures the goal, context, and requirements from the Product Owner's input.

/refine Agent

Deep codebase research. Spawns 3 parallel research agents to map affected areas, find patterns, and detect gaps. Then drives a guided Q&A conversation before writing.

/review-ticket

Validate a ticket against quality gates before marking it ready. Checks for completeness, testable requirements, and clear acceptance criteria.

Designer

Commands to shape the interface using the ticket as the source of truth.

/design

Fill the UI/UX section of a ticket with component references from the codebase. Maps existing patterns and components to the new requirements.

/review-design

Validate design specs against existing components and patterns. Flags inconsistencies or missing references before handing off to the Developer.

Developer — Blueprint Lifecycle

Commands to generate, update, and validate the technical recipe.

/scaffold

Generate a blueprint from a ticket. Pulls the structured content and creates the technical recipe in .blackbox/blueprints/.

/refresh

Update a blueprint after a send-back. Overwrites ticket-owned sections while preserving Developer-owned sections.

/refine Agent

Deep codebase research. Spawns 3 parallel research agents to map affected areas, find patterns, and detect gaps. Drives a guided Q&A before writing.

/send-back

Return a ticket to Product Owner or Designer when requirements are unclear or incomplete.

Developer — Development Flow

Commands to execute and track the actual implementation.

/start

Review the blueprint summary and prepare to begin work.

/implement

Execute the implementation plan step by step with task tracking.

/review-code

Self-review changed files against a quality checklist before committing.

/wrap-up

Clean code, run linters, and create a Pull Request document in .blackbox/prs/.

/create-pr

Create a GitHub Pull Request from the Pull Request document.

/update-pr

Update an existing GitHub Pull Request from the Pull Request document.

Developer — Review & Status

Commands to monitor progress and maintain quality.

/pr-review

Review someone else's Pull Request.

/status

List blueprints and their progress.

/export-session

Package the current session data into a compressed archive for sharing or later analysis.

/analyze-session

Parse a session export to extract factual corrections, research gaps, and protocol violations.

Developer — Project Setup

Commands to configure the project for all roles.

/setup-roles Agent

Generate personalized CLAUDE.md files for PO and DESIGN roles. Spawns parallel agents to research the codebase, then writes role-specific configurations to .blackbox/roles/.

Developer — Skill Toolkit

Commands to maintain and improve skills themselves.

/validate-skill

Validate skill files against Anthropic's guide. Checks frontmatter, naming, description triggers, required sections, and instruction quality.

Structure, Not Automation

Skills don't replace the developer. They keep Claude focused on what was agreed — the ticket, the blueprint, the plan. Every decision still goes through you.