Using Claude Code and Harness to Turn Specs into Reliable Agent Workflows

说明与用途

通过 Claude Code 和 Harness 的结构化规范和自动化工作流程，父母开发者能够有效利用碎片化工作时间，生成可靠代码并保持文档最新。

问题

When work time is fragmented by parenthood, manual coding becomes inefficient and error‑prone. Agents need clear, testable specs to function reliably, but developers often skip detailed documentation.

结果

By structuring specs with constraints and acceptance criteria, agents produce reliable, testable code that fits into fragmented work sessions, enabling the developer to maintain progress on side projects while caring for a newborn.

工作流

• 1. Ideate with /ideate – research competitors and validate idea viability.
• 2. Product planning with /product-plan – create audience, positioning, roadmap, UX direction, and risks.
• 3. Store outputs in .harness/product/ (idea.md, product.md, roadmap.md, competitors.md, CONTEXT.md).
• 4. Dev planning with /dev-plan – generate architecture, stack decisions, ADRs, and feature specs with goals, scope, approach, edge cases, and acceptance criteria.
• 5. Implement with /implement – agents read specs, classify AFK features, and produce vertical slices.
• 6. QA with /qa – run tests, API calls, CLI commands, or Playwright scripts to verify acceptance criteria.
• 7. Review manually – inspect diffs, test flows, and guide agents if needed.
• 8. Merge directly into main (no PRs for personal projects).
• 9. Update docs with /update-docs – sync README and specs with code.
• 10. Use worktrees for parallel agent work on different branches.
• 11. Repeat steps for new features or projects.

证据

Agents get much better when the work has already been turned into something testable.

来源

/implement reads the specs, classifies which features can be done AFK and which ones need human approval, and implements the current phase as vertical slices.

来源

/qa comes after that and tests those features against their acceptance criteria.

来源