Coding Orchestration
Orchestrate repo-aware coding agents from issue intake to tests and pull request drafts.
Written and reviewed by Eric Han · Co-founder & CTO, reopt
Key takeaways
- Coding orchestration turns a vague issue into scoped work, a patch, verification evidence, and a reviewable pull request draft.
- The workflow runs through intake, context, plan, implement, verify, and review stages, each with a defined output.
- Run code in an isolated Sandbox and keep each agent's write scope narrow.
- Require a final file list and command results, and never bypass human review for production changes.
- Capture recurring workflows as reusable skills or templates.
Versioned agent instructions
Next.js 16.2 bundles version-matched docs at node_modules/next/dist/docs/. Managed instructions are enabled by default in 16.3+, created by create-next-app or updated by next dev when an agent is detected. Keep team rules outside the managed block and enforce required checks through CI and branch protection.
next-devtools-mcp discovers the development server and forwards runtime tools; browser verification is a separate CLI workflow. See AI agents and MCP.
Coding orchestration should turn a vague issue into scoped work, a patch, verification evidence, and a reviewable pull request draft.
Workflow
| Stage | Output |
|---|---|
| Intake | Issue, acceptance criteria, affected area |
| Context | Repo map, commands, ownership, constraints |
| Plan | Files to inspect or change |
| Implement | Scoped patch |
| Verify | Tests, build, lint, manual checks |
| Review | Diff summary, risks, PR body |
Runtime Rules
- Use Sandbox for isolated code execution.
- Keep write scope narrow for each agent.
- Require final file list and command results.
- Never bypass human review for production changes.
- Feed recurring workflows into skills or templates.