SDD Walkthrough

Purpose: Load this when about to orchestrate an execute-plan session. Provides concrete worked examples of all five common task outcomes so you know exactly what delegate() calls to make and how to respond to each reviewer verdict.

Why This Skill Exists

The subagent-driven-development skill describes the process. This skill shows it in action — real delegate() call syntax, realistic reviewer responses, and the decisions an orchestrator makes at each step. Reference it before you start so the patterns are fresh when edge cases appear.

Scenarios Covered

Companion File

See five-task-example.md for the complete worked example — all five tasks with full delegate() calls, implementer/reviewer responses, fix loops, and a completion summary.

Quick Reference: delegate() Call Shapes

# Implementer — always fresh context, coding model
delegate(
    agent="superpowers:implementer",
    instruction="...",   # full task spec + scene-setting context
    context_depth="none",
    model_role="coding",
)

# Spec reviewer — needs recent agent results to see what was built
delegate(
    agent="superpowers:spec-reviewer",
    instruction="...",   # task spec + commit refs
    context_depth="recent",
    context_scope="agents",
)

# Code quality reviewer — same context shape as spec reviewer
delegate(
    agent="superpowers:code-quality-reviewer",
    instruction="...",   # commit refs + what was changed
    context_depth="recent",
    context_scope="agents",
)

Status Handling Cheat Sheet