Execute implementation plans by dispatching fresh subagents per task, with two-stage review after each: spec compliance review first, then code quality review. Ensures high quality and fast iteration.
- When to use decision tree (have plan? tasks independent? stay in session?)
- The process flow (dispatch implementer → spec review → code quality review → next task)
- Prompt templates (implementer, spec reviewer, code quality reviewer)
- Example workflow walkthrough
- Advantages vs manual execution and executing-plans
- Red flags and rules for subagent interaction
- Have an implementation plan with mostly independent tasks
- Want to stay in the current session (vs. parallel sessions)
- Need fresh context per task to avoid context pollution
- Want automatic quality gates (spec compliance + code quality)
- Faster iteration without human-in-loop between tasks
npx add https://github.com/wpank/ai/tree/main/skills/meta/subagent-developmentnpx clawhub@latest install subagent-developmentFrom your project root:
mkdir -p .cursor/skills
cp -r ~/.ai-skills/skills/meta/subagent-development .cursor/skills/subagent-developmentmkdir -p ~/.cursor/skills
cp -r ~/.ai-skills/skills/meta/subagent-development ~/.cursor/skills/subagent-developmentFrom your project root:
mkdir -p .claude/skills
cp -r ~/.ai-skills/skills/meta/subagent-development .claude/skills/subagent-developmentmkdir -p ~/.claude/skills
cp -r ~/.ai-skills/skills/meta/subagent-development ~/.claude/skills/subagent-developmentwriting-plans— Creates the plan this skill executesrequesting-code-review— Code review template for reviewer subagentsfinishing-a-development-branch— Complete development after all taskstest-driven-development— Subagents follow TDD for each taskexecuting-plans— Alternative for parallel session execution
Part of the Meta skill category.