Where Are the Main Configuration Files in reverse-skill?

The main configuration files in reverse-skill are RULES.md at the repository root, skills/routing.md in the skills hierarchy, and skills/scripts/bootstrap-manifest.json for environment-specific settings.

The reverse-skill repository uses a centralized configuration layer to drive AI behavior, routing logic, and tool discovery. All critical config files live at the repository root or within the skills/ subdirectory, making them easy to locate and version-control. This article maps every key file, explains its role in the execution chain, and shows how the bootstrap and routing scripts consume these configurations.

Core Configuration Files at Repository Root

The root-level files define global policies and AI-specific behavior chains.

RULES.md — The Single Source of Truth

RULES.md sits at https://github.com/zhaoxuya520/reverse-skill/blob/main/RULES.md and contains the global routing rules, execution chain definitions, scope-gate policies, and hard-coded constraints that every AI must obey. No skill executes without first loading this file.

CLAUDE.md — AI Model-Specific Behavior

CLAUDE.md (root level) defines how the Claude-Code model should interpret other configurations. It sits downstream of RULES.md in the loading order and fine-tunes response patterns for this specific AI architecture.

README_AI.md — Agent Onboarding Protocol

Located at the root, README_AI.md provides AI-focused onboarding instructions. According to the source code, it instructs agents to refresh the tool index before reading RULES.md, ensuring the environment is current before policy evaluation begins.

README.md and Translations

• README.md — High-level project overview and quick-start links
• RULES_zh.md — Chinese translation of RULES.md, functionally identical

Skills Hierarchy Configuration Files

The skills/ directory contains operational configs that bridge policy to execution.

skills/routing.md — Human-Readable Routing Table

skills/routing.md maps trigger keywords to skill scripts. The master-route.ps1 script parses this file at runtime to dispatch requests. Keywords referenced here must be defined in RULES.md to maintain coherence.

skills/tool-index.md.template — Tool Discovery Template

This template at skills/tool-index.md.template defines the structure for the generated tool-index.md/json files. Bootstrap scripts populate this template by scanning the local environment for available binaries and tools.

skills/scripts/bootstrap-manifest.json — Environment Manifest

bootstrap-manifest.json provides JSON-structured path resolution for binaries, scripts, and environment-specific directories. Both bootstrap-reverse.ps1 and bootstrap-reverse.sh consume this file during initialization.

Platform-Specific and Validation Configs

kali/RULES-kali.md — Linux-Variant Rules

For Kali Linux deployments, kali/RULES-kali.md provides identical semantics to RULES.md but with Linux-specific commands and file paths. This is the only platform-divergent config file in the repository.

Verification and Coherence Checking

skills/scripts/verify-routing-coherence.ps1 validates consistency across RULES.md, routing.md, and the generated tool-index. Run this after any configuration change to catch routing mismatches before execution.

Configuration Loading Order

The reverse-skill bootstrap sequence follows a strict chain:

1. bootstrap-reverse.ps1 (Windows) or bootstrap-reverse.sh (Linux/macOS) launches
2. Reads README_AI.md to confirm the refresh requirement
3. Executes refresh-tool-index.ps1 or refresh-tool-index.sh to populate tool-index.md/json
4. Loads RULES.md into the execution context
5. master-route.ps1 parses routing.md and dispatches to the appropriate skill

Practical Code Examples

Refresh the tool index before any skill execution:

# Windows — from repository root

.\skills\scripts\refresh-tool-index.ps1

Launch the primary entry point, which auto-loads configs:

# Windows PowerShell

.\skills\scripts\master-route.ps1

Linux/macOS equivalent workflow:

# From repository root

bash skills/scripts/refresh-tool-index.sh
bash skills/scripts/master-route.ps1  # Requires PowerShell Core (pwsh)

Validate configuration coherence after edits:

.\skills\scripts\verify-routing-coherence.ps1

Inspect the generated tool index:

Get-Content .\skills\tool-index.json | ConvertFrom-Json | Format-Table

Quick-Reference: All Main Configuration Files

File	Path	Purpose
Global rules	RULES.md at root	Routing rules, execution chain, scope gates
AI behavior	CLAUDE.md at root	Model-specific interpretation layer
Agent onboarding	README_AI.md at root	Refresh-then-read protocol
Routing table	skills/routing.md	Keyword-to-skill mappings
Tool template	skills/tool-index.md.template	Structure for discovered tools
Environment manifest	skills/scripts/bootstrap-manifest.json	Binary and path resolution
Platform rules (Kali)	kali/RULES-kali.md	Linux-specific policy variant

Summary

• RULES.md at the repository root is the mandatory starting point for all configuration queries in reverse-skill
• skills/routing.md and bootstrap-manifest.json bridge policy to executable skills
• The bootstrap sequence enforces a refresh-then-load pattern via README_AI.md instructions
• Platform parity is maintained through identical file structures, with only RULES-kali.md diverging for Linux paths
• Run verify-routing-coherence.ps1 after any configuration change to prevent routing failures

Frequently Asked Questions

What is the most important configuration file in reverse-skill?

RULES.md at the repository root is the single source of truth. Every skill execution begins with loading this file, and it defines the global routing rules, execution chains, and hard-coded policies that constrain all AI behavior.

How does reverse-skill handle different operating systems?

The repository uses identical config file structures across Windows and Linux. The only divergence is kali/RULES-kali.md, which provides Linux-specific commands and paths while maintaining the same semantic rules as the root RULES.md. Bootstrap scripts detect the platform and select the appropriate shell (PowerShell or Bash).

Why must I refresh the tool index before running skills?

According to README_AI.md and the bootstrap implementation, the tool index refresh ensures the environment-specific tool-index.json reflects currently available binaries. master-route.ps1 relies on this index for path resolution when dispatching to skills, and stale indices cause routing failures.

Where do I add new trigger keywords for skill routing?

Add keywords to skills/routing.md following the existing markdown table format. Ensure each keyword is also defined in RULES.md to maintain coherence, then run verify-routing-coherence.ps1 to validate the change before execution.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}