skills.rest
luicabref97
luicabref97COMMUNITY
Explore all skills by this author →
GitHub

sdd:plan

Refine draft task specifications into implementation-ready tasks via staged agent workflows.

Updated Apr 6, 2026
Search Tags:#decomposition#code-analysis#human-in-the-loop#workflow-orchestration#task-refinement#architecture-synthesis#llm-judges
One-click install
npx skills add https://github.com/luicabref97/sushi-jungle-web --skill sdd-plan-luicabref97
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sdd:plan
Source: https://github.com/luicabref97/sushi-jungle-web/tree/main/.agents/skills/sdd-plan
Command: npx skills add https://github.com/luicabref97/sushi-jungle-web --skill sdd-plan-luicabref97

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill automates the orchestration of multi-agent task refinement to convert ambiguous or incomplete draft task specifications into fully planned, implementation-ready tasks. It enforces staged analysis, architecture synthesis, decomposition, parallelization, and verification with judge gates and optional human checkpoints so that errors do not propagate through the workflow.

Core Features & Use Cases

  • Parallel Analysis: Launches research, codebase impact analysis, and business analysis in parallel to gather the inputs needed for design.
  • Architecture & Decomposition: Synthesizes findings into an architecture overview and breaks work into implementation steps with risks and critical path identification.
  • Verification & Promotion: Runs LLM-based judges with configurable quality thresholds and retries, supports human-in-the-loop checkpoints, and promotes refined tasks from draft to todo.
  • Incremental Refinement: Supports a --refine mode that detects git diffs to re-run only affected stages, and flags/options to continue from specific stages or skip judges for faster runs.

Quick Start

Run the plan workflow on a draft task file to produce a refined, implementation-ready task and promote it to .specs/tasks/todo/.

Frequently Asked Questions about sdd:plan

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

FAQPage Schema
How do I refine draft task specifications into implementation-ready plans?▼

To refine draft task specifications, you orchestrate staged agent workflows that run parallel research, codebase analysis, and architecture synthesis to convert drafts into implementation-ready tasks.

What is multi-agent task decomposition and how does it handle codebase analysis?▼

Multi-agent task decomposition breaks work into implementation steps by synthesizing parallel research and codebase impact analysis into an architecture overview, identifying risks and the critical path.

Can I use LLM judges to verify task refinement quality before implementation?▼

Yes, LLM-based judges with configurable quality thresholds and retries verify task refinement quality, supporting human-in-the-loop checkpoints and promoting refined tasks from draft to todo.

How do I re-run only affected workflow stages after making code changes?▼

You re-run only affected workflow stages by using a refine mode that detects git diffs, which allows you to skip unchanged stages and optionally continue from specific stages for faster execution.

Do I need git access to run workflow orchestration for task refinement?▼

Yes, you need git access and the CLAUDE_PLUGIN_ROOT environment variable set to run workflow orchestration, as git access is required for change detection during refine mode.

What is the best way to automate architecture synthesis from ambiguous draft files?▼

The best way to automate architecture synthesis from ambiguous draft files is launching parallel business analysis and research agents, then synthesizing findings into an architecture overview with risk identification.

Related Skills

Explore More Software Engineering Skills→
typemdtypemd
9

create-release

Gather closed issues and commits to draft release notes and create a GitHub draft release.

Official
Advanced
riezriez
1

subagent-driven-development

Dispatch specialized agents for independent tasks with two-stage reviews.

Community
Advanced
jxtngxjxtngx
2

langsmith-evaluator

Create and deploy LLM agent evaluation pipelines within the LangSmith ecosystem.

Community
Advanced
GamezarGamezar

sdd-plan

Refine draft task specifications into implementation-ready tasks via multi-agent workflow.

Community
Advanced