subagent-guide

To provide context to subagents with zero knowledge, include all necessary task details, constraints, and success criteria in a single comprehensive prompt before delegation, ensuring they can operate without asking clarifying questions.

Prevent flawed implementations by embedding critical thinking requirements into your prompts, adding sanity checks, domain awareness, and guardrails to guide subagents against unsafe or incorrect coding patterns.

Use worktree isolation when launching multiple agents against the same repository, preventing file conflicts and ensuring safe agent coordination by isolating each agent's working directory.

Subagent prompts should reference CODING-STANDARDS.md and the latest Opus model guidance to enforce implementation safety, ensuring subagents follow established coding conventions during task execution.

The best way to structure background agent prompts is describing the task, constraints, and success criteria comprehensively in a single prompt, applying the Critical Context Principle for safe and effective delegation.

Subagents produce poor output when prompts lack critical thinking requirements or sufficient context, as they cannot ask clarifying questions and rely entirely on the completeness of provided information for execution.