name: sdd-walkthrough description: Use when about to orchestrate a subagent-driven-development execute-plan session — provides 5 realistic task scenarios with Amplifier delegate() patterns, model_role selection, status handling (DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT), and fix loops for spec and quality issues
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
| # | Task | Outcome Pattern | Key Technique |
|---|---|---|---|
| 1 | Validate email format | Happy path: DONE → spec PASS → quality PASS | context_depth='none', model_role='coding' |
| 2 | Domain reachability check | Spec gap caught: impl raises instead of returns False | Re-delegate with specific gap description |
| 3 | In-memory rate limiter | DONE_WITH_CONCERNS: multi-worker incompatibility | Note concern, proceed to review, surface to user |
| 4 | Validation endpoint wiring | Quality issues: magic number + missing error handling | Fix loop with attempt counter in instruction |
| 5 | Error response handler | NEEDS_CONTEXT: can't find error format | Grep investigation, re-delegate with discovery |
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
| Status | Meaning | Orchestrator Action |
|---|---|---|
DONE | Task complete, tests pass | Proceed to spec review |
DONE_WITH_CONCERNS | Complete but flagged an issue | Read concern, note it, proceed to review |
NEEDS_CONTEXT | Missing info to proceed | Investigate (grep/read), re-delegate with context |
BLOCKED | Cannot complete at all | Assess blocker, provide context or escalate |