LaunchKit · 2026
Back to Skills

subagent-dev

Execute implementation plans using fresh subagents per task with two-stage review (spec compliance + code quality).

0
0 downloads
by @kjaylee

Setup & Installation

openclaw skills install @kjaylee/subagent-dev

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install subagent-dev

What This Skill Does

A development methodology that executes implementation plans by dispatching fresh subagents per task, each followed by two sequential review stages: spec compliance and code quality. Each task gets an isolated agent context to prevent cross-task contamination. Designed as an enhancement to the ralph-loop workflow.

Using one reviewer for both spec compliance and code quality tends to conflate the two concerns, causing both to be checked superficially — splitting them into sequential gates catches more issues.

When to use it

  • Breaking a multi-task feature plan into isolated agent executions
  • Catching spec drift before merging a completed task
  • Separating "did we build the right thing" from "did we build it well" reviews
  • Running TDD cycles through subagents with commit-per-task discipline
  • Handling ambiguous requirements by letting subagents ask clarifying questions mid-task

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: Execute task 3 from the implementation plan: add user authentication middleware

AGENT
  1. 1Extract task 3 spec and relevant context (files, architecture constraints) from the plan
  2. 2Dispatch implementation subagent with task spec, context, TDD constraint, and self-review requirement
  3. 3Implementation subagent writes tests first, implements, commits, and returns summary with self-review findings
  4. 4Dispatch spec-review subagent with original requirements and git diff to verify all requirements are met and no extras added
  5. 5Dispatch code-quality subagent to check naming, duplication, error handling, and edge cases — loop back to implementer if issues found
OUTPUT

Task 3 marked complete with two review gates passed, commit made, ready for next task