2.5k
GitHub Stars
2
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill openclaw/skills --skill subagent-dev- _meta.json277 B
- SKILL.md3.1 KB
Overview
This skill executes implementation plans by dispatching fresh subagents per task and enforcing a two-stage review (spec compliance and code quality). It is designed to run the implement phase of a ralph-loop style workflow with isolated, repeatable work units. The approach improves correctness and maintainability at the cost of additional subagent runs and controller preparation.
How this skill works
The controller reads the plan and extracts tasks. For each task, it launches a fresh implementer subagent that follows TDD, produces commits, runs tests, and generates a self-review. Then a spec-review subagent verifies requirements match the implementation, and a quality-review subagent inspects code hygiene, edge cases, and security. Tasks repeat until both review stages approve, then the branch undergoes a final review and closure.
When to use it
- Implementing multi-task plans where tasks can be executed independently
- Projects requiring high confidence in requirement fidelity and code quality
- When you want reproducible, uncontaminated task context for each unit of work
- Feature work with complex acceptance criteria or many edge cases
- Critical systems where defects are costly and extra review is justified
Best practices
- Prepare controller context: concise task description, relevant files, and constraints before dispatch
- Keep each task scope minimal so subagents focus and converge faster
- Enforce TDD in the implementer prompt: write tests first, then implementation and commits
- Require clear commit messages and a brief self-review from implementer subagents
- Automate retry loops for reviewer findings so the implementer receives actionable feedback
Example use cases
- Breaking a large feature into isolated tasks and ensuring each meets spec and style
- Onboarding a team by encoding standards into review subagent prompts for consistent feedback
- Refactoring modules in small increments with a guarantee that behavior remains unchanged
- Implementing security-sensitive features where both correctness and quality reviews are mandatory
FAQ
Separating spec verification from code quality reduces cognitive load and prevents tradeoffs where one concern masks the other. Spec review ensures the implementation matches requirements; quality review ensures it is robust and maintainable.
What is the cost in subagent calls?
Typically three subagent runs per task (implementer, spec reviewer, quality reviewer), plus possible re-dispatches when questions or fixes are required.