paralelizacao-em-grafo

Decides when to split work into parallel agent slices and orchestrates fan-out and fan-in execution.

1|Updated Jun 2, 2026
One-click install
npx skills add https://github.com/psiagoleal/ai-coding-agent-profiles --skill paralelizacao-em-grafo-psiagoleal
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: paralelizacao-em-grafo
Source: https://github.com/psiagoleal/ai-coding-agent-profiles/tree/main/skills/paralelizacao-em-grafo
Command: npx skills add https://github.com/psiagoleal/ai-coding-agent-profiles --skill paralelizacao-em-grafo-psiagoleal

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Parallelizing work across multiple AI agents often backfires: slices share files, contracts stay open, and the merge becomes a costly negotiation that erases any time saved. This Skill provides the decision criteria and execution protocol to know when a parallel graph actually pays off and how to run it without rework. ## Core Features & Use Cases - Go/no-go decision framework: Enforces a closed contract (schema, types, error format) before any fan-out, and requires vertical slices with zero shared files. - Isolation protocol: Assigns each slice its own git worktree with a declared arquivos_exclusivos manifest verified against all other slices in the wave. - Single-owner fan-in: Defines an integrator role that only merges and checks integration seams, never implements, and rejects failed slices back to their worktrees. - Cost awareness: Explains the clock-vs-token tradeoff (N× prefill against subscription quota) and lists cases where parallelizing is pure overhead. - Use Case: When planning a wave with multiple tasks — e.g., building auth and export features simultaneously — use this Skill to verify the slices are truly independent, spin up worktrees, and run a clean single-owner integration. ## Quick Start Ask the agent to evaluate whether your planned multi-task wave qualifies for parallel execution and to set up the worktree-based fan-out and fan-in plan.

Frequently Asked Questions about paralelizacao-em-grafo

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

FAQPage Schema
How do I parallelize coding tasks across multiple AI agents?

First close the contract (schema, types, error format), then split work into vertical slices where no file is read or written by two slices. Run each slice in its own git worktree with its own verification gates, and finish with a single-owner fan-in.

When should I not parallelize work between agents?

Avoid parallelizing single-task waves, slices sharing files, work with an open contract, tasks depending on conversation history, or when token cost matters more than deadline. In these cases sequential execution or one owner for the coupled set is cheaper.

What is the difference between vertical and horizontal task slices?

A vertical slice gives one agent an entire feature (schema, API, UI) with exclusive files, so pieces join cleanly at fan-in. A horizontal slice splits by layer (models, routes, types), forcing agents to share files and turning the merge into a negotiation.

How do git worktrees help run agents in parallel?

git worktree gives each agent a separate checkout of the same repository, preventing conflicts or silent overwrites when slices edit concurrently. Create one worktree per slice with its own branch, then remove the temporary worktrees after fan-in.

Why does parallel agent execution consume so much quota?

Each parallel window re-reads the contract and codebase from scratch, so N slices cost N times the prefill tokens, drawn from the same subscription quota. The clock time saved is only the slowest slice, while the bill multiplies.

What should the integrator do during fan-in of parallel work?

The integrator reviews all slice outputs in one window, checks integration seams like cross-layer mismatches, and signs off a single result. The integrator must not implement code, and failed slices are returned to their worktrees rather than merged.