How does this skill prevent agents from overwriting each other's work?

It enforces a file reservation system where agents must reserve a file using a TTL (Time To Live) before making edits and release it upon completion.

How are subtasks managed in this framework?

Whenever a swarm subtask is spawned, the skill ensures an immediate follow-up call to initialize a worker agent with the specific prompt returned by the swarm system.

What is the primary role of Always-On Guidance?

It provides a set of persistent rules that govern how Claude agents interact with tools, manage file edits, and coordinate tasks within a swarm architecture.

Does this skill work with different AI models?

Yes, it includes specific guidance for different model tiers, such as GPT-5.2-code for concise imperatives and Opus 4.5 for multi-phase rationale.

What testing standards does the skill enforce?

It mandates a red-green-refactor workflow whenever tests are available for the code being modified, ensuring high code quality.

Always-On Swarm Guidance

Name: Always-On Swarm Guidance
Author: primeinc

byprimeinc

0•

Productivity & Workflow

Enforces standardized agent behaviors, model-specific defaults, and swarm coordination rules for multi-agent development workflows.

Always-On Swarm Guidance provides a foundational framework for Claude agents operating within complex multi-agent environments. It establishes a clear hierarchy of instruction priority, mandates the use of specific swarm tools, and manages resource locking through file reservations to prevent conflicts during concurrent edits. By defining distinct operational modes for different model tiers—emphasizing concise imperatives for code-specific models and reasoned rationales for larger models—this skill ensures consistent, high-performance output across distributed tasks while enforcing a strict red-green-refactor testing discipline.

Key Features

01Instruction priority enforcement (System > Developer > User > Agents)

02Model-specific behavior tuning for GPT-5.2-code and Opus 4.5

030 GitHub stars

04Red-Green-Refactor testing discipline integration

05Automated file reservation and TTL-based locking mechanisms

06Standardized swarm task spawning and sub-agent initialization

Use Cases

01Coordinating multiple Claude agents working on a shared repository to prevent file-write collisions.

02Standardizing AI output formats across different model backends for predictable script parsing.

03Ensuring automated development tasks follow a strict test-driven development (TDD) workflow.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add primeinc/swarm always-on-guidance

For use in Claude.ai and ChatGPT

Download Skill