How does the gate system work?

Hard gates (subsystem invariants, tier0 tests) must pass for work to be committed. Soft gates (tier1 tests, documentation coverage) provide warnings but allow the process to continue unless too many accumulate.

When should I use the Harness Protocol?

Use this skill when running autonomous engineering loops where Claude needs to make independent decisions about proceeding, reverting, or pausing work without human supervision.

How does it handle context limits?

It includes specific triggers to save state, commit existing progress, and pause the harness when context usage reaches critical levels to maintain high reasoning quality.

What happens if a task fails?

The protocol analyzes if a failure is retryable (like a test flake) or structural. It will attempt one alternate approach for retryable errors before logging a detailed failure report and skipping to the next task.

Does this skill help with git management?

Yes, it enforces a strict 'one commit per change group' rule with structured messages, ensuring a clean history and making it easy to revert specific failed attempts.

Harness Protocol

Name: Harness Protocol
Author: mberto10

bymberto10

•

生産性とワークフロー

Establishes a rigorous decision-making framework for autonomous engineering cycles, including gate checks, failure handling, and commit discipline.

The Harness Protocol skill provides a sophisticated framework for managing autonomous engineering tasks within Claude Code. It defines the rules for unsupervised plan-work-review-commit cycles, ensuring that AI-driven changes are validated against 'Hard Gates' like subsystem invariants and 'Soft Gates' like spec coverage. By categorizing failures into retryable or structural types and enforcing strict commit discipline, this skill allows for reliable, high-integrity code generation and automated project management via tools like Linear without requiring constant human intervention.

主な機能

01Automated Hard/Soft Gate verification for code quality

02Context budget management for stable long-running tasks

031 GitHub stars

04Advanced failure taxonomy with intelligent retry logic

05Structured commit discipline for granular reverts

06Automated Linear issue parsing and status management

ユースケース

01Automating bug fixes and feature implementation from issue trackers

02Enforcing repository-wide invariants and testing standards during AI development

03Running unsupervised autonomous engineering cycles with /harness start

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mberto10/mberto-compound harness-protocol

For use in Claude.ai and ChatGPT

主な機能

01Automated Hard/Soft Gate verification for code quality

02Context budget management for stable long-running tasks

031 GitHub stars

04Advanced failure taxonomy with intelligent retry logic

05Structured commit discipline for granular reverts

06Automated Linear issue parsing and status management

ユースケース

01Automating bug fixes and feature implementation from issue trackers

02Enforcing repository-wide invariants and testing standards during AI development

03Running unsupervised autonomous engineering cycles with /harness start

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mberto10/mberto-compound harness-protocol

For use in Claude.ai and ChatGPT