---
name: feature-build
description: Recon → plan → implement → parallel review → synthesize. Five-stage workflow for shipping features end-to-end with Ollama Cloud subagents.
---

## scout
phase: Context
label: Map codebase
as: context
output: context.md
progress: true

Map the relevant code for {task}. List files to touch, capture conventions, identify risks, and write findings to {output}.

## ollama-planner
phase: Planning
label: Create implementation plan
as: plan
reads: context.md
output: plan.md
progress: true

Read {reads} and create a concrete implementation plan for {task}. Write ordered tasks with file paths, acceptance criteria, and risks to {output}.

## ollama-coder
phase: Implementation
label: Implement the feature
as: implementation
reads: plan.md
output: implementation.md
progress: true

Execute the plan from {reads}. Implement the smallest correct changes, run validation, and report changed files with evidence to {output}.

## reviewer
phase: Review
label: Correctness review
as: correctness_review
reads: plan.md, implementation.md
output: review-correctness.md
progress: true

Review the implementation against the plan. Focus on: correctness, regressions, edge cases, and whether the change does what it claims. Write findings to {output}.

## reviewer
phase: Review
label: Tests and validation review
as: test_review
reads: plan.md, implementation.md
output: review-tests.md
progress: true

Review test coverage and validation. Focus on: tests added/updated, do they exercise the new code, are they sufficient to catch regressions. Write findings to {output}.

## reviewer
phase: Review
label: Simplicity and security review
as: simplicity_review
reads: plan.md, implementation.md
output: review-simplicity.md
progress: true

Review for simplicity, maintainability, and security. Focus on: smallest correct change, readability, naming consistency, input validation, secrets, injection risks. Write findings to {output}.

## worker
phase: Synthesis
label: Synthesize and apply fixes
reads: review-correctness.md, review-tests.md, review-simplicity.md
output: synthesis.md
progress: true

Synthesize the three reviews from {reads}. Apply fixes worth doing now (small, safe changes). Report: what was fixed, what was deferred, and any blockers that need user decision.