anti-drift-swarm

Enforces coordinator checkpoints and per-wave caps for parallel subagent dispatch.

Updated Apr 15, 2026
One-click install
npx skills add https://github.com/ToanPV90/dotfiles --skill anti-drift-swarm
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: anti-drift-swarm
Source: https://github.com/ToanPV90/dotfiles/tree/main/claude/.claude/skills/anti-drift-swarm
Command: npx skills add https://github.com/ToanPV90/dotfiles --skill anti-drift-swarm

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Anti-drift-swarm prevents multi-agent dispatch from going off-track by enforcing strict coordination, wave sizing, and return checkpoints so work remains reconcilable and goal-aligned.

Core Features & Use Cases

  • Wave-based coordination discipline: dispatches subagents in controlled waves with a coordinator checkpoint between waves to stop drift from compounding.
  • Role specialization with single deliverables: requires each subagent to have one deliverable and one file scope, avoiding mixed research-and-implementation prompts that cause shallow outputs.
  • Safety guardrails for file-writing subagents: forces write-capable subagents to run in the foreground and pre-creates target directories to prevent silent failures and race conditions.
  • Practical integration fit: designed to complement parallel dispatch and subagent-driven development skills by specifying the “how” and “when” at the protocol level.

Quick Start

Use anti-drift-swarm before dispatching a parallel wave of three or more subagents that will read from and/or write to multiple files, especially when you need reliable reconciliation between waves.

Frequently Asked Questions about anti-drift-swarm

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

FAQPage Schema
How do I prevent goal drift when dispatching multiple subagents in parallel?

To prevent goal drift during parallel subagent dispatch, enforce a coordinator-only reconciliation pattern with strict per-wave parallel caps and required checkpoint protocols to keep work goal-aligned.

Why do my parallel subagents produce shallow outputs during multi-file refactors?

Parallel subagents produce shallow outputs when given mixed research-and-implementation prompts. Assigning each subagent a strict single deliverable and one file scope prevents shallow outputs during multi-file refactors.

How do I avoid race conditions with write-capable subagents writing to multiple files?

To avoid race conditions with write-capable subagents, force them to run in the foreground and pre-create target directories before dispatch, which prevents silent failures during parallel file writes.

When should I use wave-based coordination for multi-agent dispatch?

Use wave-based coordination for multi-agent dispatch before dispatching a parallel wave of three or more subagents that read from or write to multiple files, ensuring reliable reconciliation between waves.

What is the best way to reconcile multiple subagents returning from a parallel research task?

The best way to reconcile multiple subagents returning from parallel research is to apply a coordinator-only reconciliation pattern, enforcing a checkpoint and restated-goal protocol between dispatch waves.

Do I need to pre-create target directories before parallel subagent dispatch?

Pre-creating target directories before parallel subagent dispatch is required to ensure write-safety. This prevents silent failures and race conditions when multiple write-capable subagents execute concurrently.