Subagent-Driven Development
Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.
Core principle: Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration
When to Use
Use when:
-
Have implementation plan
-
Tasks are mostly independent
-
Want to stay in current session
-
Want fast iteration with review checkpoints
vs. Executing Plans (parallel session):
-
Same session (no context switch)
-
Fresh subagent per task (no context pollution)
-
Two-stage review after each task: spec compliance first, then code quality
-
Faster iteration (no human-in-loop between tasks)
The Process
Setup
-
Read plan, extract all tasks with full text and context
-
Create TodoWrite with all tasks
Per Task
-
Dispatch implementer subagent with full task text + context
-
If subagent asks questions - Answer, provide context
-
Implementer implements, tests, commits, self-reviews
-
Dispatch spec reviewer subagent - Verify code matches spec
-
If spec issues - Implementer fixes, reviewer re-reviews
-
Dispatch code quality reviewer subagent - Review for quality
-
If quality issues - Implementer fixes, reviewer re-reviews
-
Mark task complete in TodoWrite
After All Tasks
-
Dispatch final code reviewer for entire implementation
-
Use ltk:finishing-a-development-branch to complete
Two-Stage Review
Stage 1: Spec Compliance
-
Does implementation match spec EXACTLY?
-
Nothing missing?
-
Nothing extra (over-building)?
Stage 2: Code Quality (only after spec passes)
-
Clean code?
-
Good test coverage?
-
Maintainable?
Advantages
vs. Manual execution:
-
Subagents follow TDD naturally
-
Fresh context per task (no confusion)
-
Parallel-safe (subagents don't interfere)
-
Subagent can ask questions (before AND during work)
Quality gates:
-
Self-review catches issues before handoff
-
Two-stage review: spec compliance, then code quality
-
Review loops ensure fixes actually work
Red Flags
Never:
-
Skip reviews (spec compliance OR code quality)
-
Proceed with unfixed issues
-
Dispatch multiple implementation subagents in parallel (conflicts)
-
Make subagent read plan file (provide full text instead)
-
Skip scene-setting context
-
Ignore subagent questions
-
Accept "close enough" on spec compliance
-
Skip review loops
-
Start code quality review before spec compliance passes
-
Move to next task while either review has open issues
If subagent asks questions:
-
Answer clearly and completely
-
Provide additional context if needed
-
Don't rush them into implementation
If reviewer finds issues:
-
Implementer (same subagent) fixes them
-
Reviewer reviews again
-
Repeat until approved
Integration
Required workflow skills:
-
ltk:writing-plans - Creates the plan this skill executes
-
ltk:requesting-code-review - Code review template for reviewer subagents
-
ltk:finishing-a-development-branch - Complete development after all tasks
Subagents should use:
- ltk:test-driven-development - Subagents follow TDD for each task
Alternative workflow:
- ltk:executing-plans - Use for parallel session instead of same-session execution