subagent-driven-development

Dispatching fresh subagents per task prevents context leakage and promotes isolated testing. This approach assigns each implementation task to a new subagent, ensuring execution remains cleanly separated and free from prior session interference.

A two-stage review in plan orchestration enforces quality by dispatching a spec-compliance reviewer subagent followed by a code-quality reviewer subagent after each task implementation. This structured sequence ensures both functional correctness and code maintainability.

Epic-based workflow management organizes related tasks and coordinates progress across parallel sessions. You can audit decisions, update task labels, and close tasks as they complete, unpinning the epic only when all associated tasks are finished.

Subagent-driven development requires no specific external dependencies. It relies on a SKILL.md frontmatter with name and description, and optionally loads scripts, references, or assets directories on demand to support task execution and orchestration.

You should use independent subagents for code implementation when executing plans with isolated tasks or coordinating parallel sessions. This method is ideal for preventing context bleed and ensuring each task undergoes strict spec and code-quality reviews.

The best way to coordinate epic workflow progress is dispatching implementer subagents for each task using prompt templates. After implementation, run two-stage reviews, update task labels, and close tasks incrementally until the entire epic is complete.