solo-code-harness

name	solo-code-harness
description	Deploy, configure, and maintain the Solo-Code multi-agent harness across Copilot, Claude, Gemini, and Kilo. Use when: deploy harness, setup copilot, setup claude, setup gemini, setup kilo, configure AI agents, install AI harness, solo-code setup, initialize project.
license	MIT

Solo-Code Harness — Multi-Agent AI Coding Framework

You are an expert on the Solo-Code Harness framework. This project provides discipline, safety, and consistency rules for AI coding agents across four platforms: GitHub Copilot (VS Code), Claude Code, Gemini, and Kilo.

---

Architecture

The harness lives in four platform-specific directories at the project root:

Platform	Directory	Auto-loads
GitHub Copilot / Cursor	.github/	copilot-instructions.md + PreToolUse hooks
Claude Code	.claude/	CLAUDE.md
Gemini	.gemini/antigravity/	AGENTS.md + skills
Kilo	.kilo/ + kilo.json + AGENTS.md	AGENTS.md, command/, agent/, skill/

Each harness includes: rulebook, skills (8-9), agents (2), memory, automation scripts.

---

Deployment

Quick Deploy (copy-to-project)

Copy this entire repository's root contents (excluding .git/, .github/workflows/, etc.) into a target project. Each AI tool auto-discovers its respective config directory.

Deploy Kilo harness only

Copy to target project:

AGENTS.md
kilo.json
.kilo/

Deploy Copilot harness only

Copy to target project:

.github/
.vscode/

---

Kilo Harness Components

The .kilo/ directory is auto-discovered by Kilo. No additional config needed.

Agents (.kilo/agent/)

Agent	Mode	Purpose
orchestrator.md	subagent	Breaks down complex tasks, coordinates parallel sub-agents
solo-code-engineer.md	subagent	Template agent for project customization

Commands (.kilo/command/)

Command	Purpose
/plan	Implementation planning with affected files, tasks, risks
/verify	Prove code works by building, running, and testing
/brainstorm	Structured idea exploration with 3+ options
/debug	Evidence-based root cause analysis before fixes
/test	Test generation and execution (AAA pattern)
/create	Feature scaffolding from scratch
/enhance	Safe refactoring of existing code
/remember	Save project conventions to persistent memory

Skills (.kilo/skill/)

Skill	When to load
code-review-expert	Reviewing code, PRs, diffs; security audits
file-editor-pro	Editing files, refactoring, precise changes
git-workflow-master	Committing, pushing, PRs, git safety
permission-guard	Destructive operations, credentials, config
systematic-debugging	Debugging, errors, unexpected behavior
brainstorming	Designing features, architecture decisions
testing-patterns	Writing tests, TDD, coverage
api-patterns	API design (REST/GraphQL/tRPC)

---

Customization Guide

For a new project

1. Copy the harness into your project
2. Customize .kilo/agent/solo-code-engineer.md with your project's tech stack, architecture map, and conventions
3. Update .kilo/memory/project-conventions.md with your project's specific rules
4. Adjust kilo.json permissions if your project needs different bash command access

Adding a new skill

Create .kilo/skill/<name>/SKILL.md with YAML frontmatter:

name: your-skill-name
description: "When to use this skill"

The skill body can use any markdown format Kilo supports.

Adding a new command

Create .kilo/command/<name>.md with YAML frontmatter:

description: What the command does
subtask: true

The command name is the filename (minus .md), invoked via /name.

---

Security Model (Defense in Depth)

The harness uses three layers of protection:

1. Permission layer (kilo.json) — Deny patterns evaluated before allow patterns. Destructive git/rm/DB/Windows-PowerShell operations blocked at the bash level.
2. Behavioral layer (AGENTS.md) — Rules instructing the AI: no destructive commands without permission, scan for secrets before commit, no runtime bypass of bash permissions.
3. Enforcement layer (skills) — permission-guard halts execution for dangerous operations; git-workflow-master enforces secret scanning with security_scan.py before commits.

---

Cross-Harness Consistency

When editing shared concepts (behavior rules, security rules, project conventions), update ALL four harnesses:

• .github/copilot-instructions.md
• .claude/CLAUDE.md
• .gemini/antigravity/AGENTS.md
• AGENTS.md (Kilo root)