l3vram

Flow Dev Company — Development skill for Claude Code

Development community

Orquestador de desarrollo end-to-end para Claude Code: enrich, plan, y flota de agentes en paralelo por olas con tiering de modelos y gates.

How to install Flow Dev Company

This entry records only its repository, not the path inside it, so there is no exact command to give. Open l3vram/flow-dev-company and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Flow Dev Company does

Orquestador de desarrollo end-to-end para Claude Code: enrich, plan, y flota de agentes en paralelo por olas con tiering de modelos y gates.

Alternatives in Development

Flow Parallel — Decompose and execute large changes, migrations, or multi-issue fixes in parallel with quality gates 2.8k ★
Agent Cost — Coste estimado de uso de agentes por sprint/proyecto 48 ★
Recursive Research — Claude Code skill for recursive research up to PhD level across any domain 41 ★

README

flow-dev-company

**End-to-end software development orchestrator for Claude Code.** One skill: give it a goal ("I want to build a chatbot"), it enriches the spec with senior-level questions, plans the work, and executes with a **fleet of parallel agents** — an expensive brain plans and reviews, cheap hands execute in isolated worktrees, every task at the right model tier.

Self-contained: **nothing else to install.**

The idea in one line

A prompter asks a question. An architect draws a graph. This skill turns "do A, then B, then C" into waves of agents running in parallel, with human gates where decisions matter and verification where trust matters.

How it works

0 Start/Resume → 1 Enrich → 2 Plan → 🚦GATE A → 3 Execute (waves) → 4 Verify → 5 Review → 🚦GATE B → 6 Handoff
                 (questions  (advisor   human    (parallel fleet,    (tests +   (branch   human   (docs)
                  → SPEC.md)  brain)             integrated per     smoke run) review)
                                                  wave)

Resume / execute mode — if a run is already in flight (.flow/state.json), it picks up where it left off. Point it at an existing plan (run plans/156-*.md) and it skips the spec and goes straight to reconcile → approval → fleet.
Enrich — specializes into your domain, proposes what good engineering includes even if unasked, and asks the missing questions with options + presets (MVP / Production / Custom). The result is written to plans/SPEC.md, including the smoke scenario that proves it works.
Plan — an advisor brain (expensive tier) writes self-contained plans plus their dependency graph (DAG).
🚦 Gate A — you approve the plan before any budget goes to the fleet.
Wave execution — the DAG is layered into waves (topological, via Kahn); each wave runs in parallel in isolated worktrees, every agent with its role (backend/frontend/data/qa/ docs…) and tier. Approved w