gds-quick-dev

Implements code changes through a spec-driven plan, implement, review, and present workflow.

Updated Sep 3, 2026
One-click install
npx skills add https://github.com/watchthelight/shatterfish --skill gds-quick-dev-watchthelight
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gds-quick-dev
Source: https://github.com/watchthelight/shatterfish/tree/main/.claude/skills/gds-quick-dev
Command: npx skills add https://github.com/watchthelight/shatterfish --skill gds-quick-dev-watchthelight

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a feature request or bug fix into reviewed, mergeable code often skips planning and review discipline. This Skill enforces a structured workflow that converts user intent into a hardened, reviewable code artifact following the project's existing architecture and conventions. ## Core Features & Use Cases - Spec-Driven Planning: Clarifies intent, investigates the codebase, and writes a scoped specification (900–1600 tokens) with tasks, acceptance criteria, and boundaries before any code changes. - Multi-Agent Adversarial Review: Runs blind, edge-case, and acceptance-auditor review subagents on the diff, classifying findings as intent gaps, spec flaws, patches, or deferred work with loopback handling. - Sprint & Epic Integration: Syncs story status in sprint-status.yaml, compiles epic context from planning artifacts, and supports both full plan-code-review and one-shot routes. - Use Case: Ask it to implement a bug fix; it drafts a spec, gets your approval, implements via sub-agent, runs three independent reviews, commits, and opens a clickable Suggested Review Order in VS Code. ## Quick Start Ask the agent to implement your feature or bug fix using the gds-quick-dev workflow and it will plan, implement, review, and present the change.

Frequently Asked Questions about gds-quick-dev

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a feature with a spec-driven AI workflow?

Invoke the skill with your intent; it clarifies requirements, investigates the codebase, writes a spec with tasks and acceptance criteria, then waits for your approval before implementing. After implementation it runs adversarial reviews and presents a clickable review order.

What is the difference between the one-shot and plan-code-review routes?

One-shot handles zero-blast-radius changes with clear intent, implementing directly with a single adversarial review. Plan-code-review covers everything else with full spec drafting, human approval, implementation, and three-reviewer classification.

Does this workflow work without sub-agent support?

Yes. Runtimes without sub-agents (e.g., Copilot, Codex, local models) fall back to inline execution for epic context compilation and generate review prompt files for the human to run in separate sessions.

How does the review step classify findings?

Findings are deduplicated and classified as intent_gap, bad_spec, patch, defer, or reject. Intent gaps and bad specs trigger loopbacks that revert code and re-derive it, while patches are auto-fixed and deferred items go to a deferred-work file.

When should I not use this workflow for a code change?

Avoid it for trivial edits where spec overhead adds no value, or when the intent is still unclear and cannot be clarified. The workflow halts at checkpoints and requires human input, so fully autonomous unattended runs are not supported.