dev-flow

Orchestrates design-to-PR delivery as isolated phase subagents with per-phase model routing.

8|15|Updated Jan 17, 2025
One-click install
npx skills add https://github.com/Netcracker/qubership-envgene --skill dev-flow-netcracker
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dev-flow
Source: https://github.com/Netcracker/qubership-envgene/tree/main/.claude/skills/dev-flow
Command: npx skills add https://github.com/Netcracker/qubership-envgene --skill dev-flow-netcracker

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Running an entire design-to-PR delivery flow in one growing conversation is expensive and fragile: context is re-read every turn, and work cannot resume after an interruption. This Skill splits the flow into isolated phase subagents that hand work over through files, keeping the orchestrator context small and each phase restartable. ## Core Features & Use Cases - Phase isolation with file hand-over: Each phase (cr, plan, implement, review, verify) runs as a fresh subagent that reads entry artifacts, invokes a child skill, writes an exit artifact, and returns only a short status. - Per-phase model routing: Mechanical work like CI polling runs on cheap models while load-bearing judgement like the final whole-branch review runs on the most capable model, with escalation rules recorded in the ledger. - Resumable flow ledger: A git-ignored ledger at .superpowers/flow/<slug>.md tracks phase completion, rulings, and model escalations so the flow resumes from the first unchecked phase after any interruption. - Use Case: A developer with a settled ADR asks to run the dev flow; the orchestrator files the CR via design-to-cr, generates a plan, drives subagent-driven implementation, runs code review, polls CI, and gates each phase for human approval. ## Quick Start Ask the assistant to run the dev flow for your settled design, naming the ADR path and doc PR so it can create the ledger and dispatch the cr phase.

Frequently Asked Questions about dev-flow

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

FAQPage Schema
How do I orchestrate a design-to-PR workflow with AI subagents?

Run the dev flow starting from a settled design: the orchestrator creates a ledger, then dispatches one phase subagent at a time (cr, plan, implement, review, verify). Each subagent reads its entry artifact files, invokes the right child skill, writes an exit artifact, and returns only a short status.

How do I resume an interrupted multi-phase development flow?

Read the flow ledger at .superpowers/flow/<slug>.md and resume at the first unchecked phase. Verify the previous phase's exit artifact actually exists in git or GitHub before dispatching, since the ledger is a claim and git is the truth.

Which model should run each phase of an AI coding workflow?

Route mechanical work like CI polling to Haiku, structured authoring and coordination like cr, plan, and the implement controller to Sonnet, and the final whole-branch review to the most capable model. Escalate when an entry artifact leaves a fork unsettled or a fix loop repeats.

When should I not use the dev-flow orchestration skill?

Do not use it to author a design from scratch, write docs, or file a CR directly; it wraps brainstorming, writing-docs, and design-to-cr rather than replacing them. The flow requires a settled design with an ADR and doc PR as its entry point.

Why hand work between agents as files instead of conversation history?

File hand-over keeps each phase restartable from nothing: a fresh subagent reads only its entry artifacts, so the orchestrator context stays small and cheap. State lives in git, the CR issue, and the ledger rather than in a growing transcript.