/sprint — Structured Squad Sprint

Use /sprint when a task spans more than one agent or involves enough complexity that planning, red-teaming, and execution need to be separated. Paulien and Bob run simultaneously (halving plan time), then Kryptonite gates before any build starts.

For a single-file or single-agent task, skip /sprint and work directly with the agent.

Note: The Workflow tool is not available in Claude Code. Parallelism is achieved with multiple Agent tool calls in a single response.

Step 1 — Confirm the brief

Before running, confirm with Steve:

• What is being built? One clear sentence.
• What does done look like? Acceptance criteria.
• Project context? Read the project CLAUDE.md and note anything the agents need to know.

Step 2 — Run the workflow

Type /sprint at the prompt. Claude Code discovers the script from ~/.claude/workflows/sprint.js automatically.

Pass args when prompted:

• brief — what is being built (one sentence)
• acceptanceCriteria — what done looks like
• projectContext — key facts from CLAUDE.md or project decisions

The workflow runs Paulien and Bob in parallel via pipeline(), then gates with Kryptonite. Watch progress in /workflows.

Step 3 — Read the Kryptonite verdict

Check result.redTeam.verdict:

• BLOCKED — stop, surface blockers to Steve, resolve before proceeding
• BUILD_WITH_CONDITIONS — proceed, but address warnings before shipping
• CLEAR_TO_BUILD — proceed

Step 4 — Execute task briefs

Hand each task to its assigned agent in dependency order. One at a time unless tasks have no dependencies.

Agent assignment key:

• Pixel — React, TypeScript, Tailwind, UI components
• Diesel — Python, Node, REST APIs, backend logic
• Atlas — Neo4j, PostGIS, data pipelines, ETL
• Orbit — Azure, CI/CD, Docker, infrastructure
• Eric — general code, anything that doesn't fit a specialist

Step 5 — Vera sign-off and /deliver

When all tasks are done, trigger Vera explicitly. Then run /deliver. Do not skip either.