How Priority Ordering Works in reverse-skill routing.json: A Complete Guide

Priority ordering in reverse-skill works as a two-step tie-breaker system: first scoring routes by keyword matches, then using the priority array to select the PRIMARY route when multiple routes have equal hit counts.

The reverse-skill project (available at zhaoxuya520/reverse-skill) implements a deterministic routing mechanism that decides which reverse-engineering skill executes first based on user hints. The entire logic lives in skills/config/routing.json, where authors control precedence through an explicit, ordered list of route IDs.

Understanding the Two-Step Routing Algorithm

Step 1: Keyword Scoring Builds the Candidate Set

Every route (R1 through R40) contains keyword rules with must patterns. When a user hint matches any pattern, that route receives one hit. After evaluating all routes, the system creates a candidate set containing only routes with at least one hit.

The scoring defaults to +1 per matched rule, though individual rules can specify custom scores. The meta.scoring section in routing.json (lines 4-7) documents this behavior and defines the fallback mechanism.

Step 2: Priority Array Resolves Ties

When multiple routes share the highest hit count, the router consults the priority array (lines 308-313) to break the tie. This array lists route IDs from highest to lowest precedence. The first ID in the array that also appears in the candidate set becomes the PRIMARY route.

This design gives authors precise, deterministic control over routing decisions without complex scoring mathematics.

The Priority Array: Author-Controlled Precedence

The priority array in routing.json explicitly orders 41 route IDs:

{
  "priority": [
    "R4", "R1", "R2", "R3", "R30", "R31", "R33", "R5", "R9", "R21",
    "R22", "R6", "R7", "R8", "R34", "R28", "R17", "R16", "R18", "R24",
    "R37", "R23", "R35", "R25", "R36", "R29", "R38", "R32", "R26", "R27",
    "R10", "R11", "R12", "R13", "R14", "R15", "R19", "R40", "R20", "R39", "R0"
  ]
}

• Position 0-9: Core reverse-engineering skills (APK analysis, binary diff, firmware extraction)
• Position 10-29: Specialized tools and frameworks
• Position 30-40: Fallback and auxiliary routes, with R0 as the ultimate generic fallback

Changing the order of IDs directly changes routing outcomes for ambiguous hints. The verify-routing-coherence.ps1 script enforces consistency between this array and the human-readable table in skills/MASTER-ROUTING.md (lines 52-96).

How the Fallback Mechanism Works

If no route receives any keyword hit, the router uses meta.fallbackId (line 5):

{
  "meta": {
    "fallbackId": "R0",
    "scoring": "Summation of matched keyword scores; priority tie-breaker"
  }
}

R0 represents the generic reverse-engineering skill. The system also suggests opening routing.md for manual review when this occurs.

Running the Router: Code Examples

Execute the Official Router Script

powershell -File skills/scripts/master-route.ps1 -Hint "analyze Android APK with Frida"

This script implements the algorithm exactly as defined in routing.json.

Simplified Algorithm Implementation

# Load routing configuration

$routes     = Get-Content skills/config/routing.json | ConvertFrom-Json
$candidates = @()

# Phase 1: Score all routes by keyword matches

foreach ($id in $routes.routes.Keys) {
    foreach ($kw in $routes.routes[$id].keywords) {
        $matchesMust    = $Hint -match $kw.must
        $passesExclude  = !($kw.exclude) -or ($Hint -notmatch $kw.exclude)
        
        if ($matchesMust -and $passesExclude) {
            $candidates += $id
            break   # One hit per route sufficient for basic scoring

        }
    }
}

# Phase 2: Resolve PRIMARY via priority ordering

if ($candidates.Count -eq 0) {
    $primary = $routes.meta.fallbackId   # "R0"

} else {
    foreach ($p in $routes.priority) {
        if ($candidates -contains $p) {
            $primary = $p
            break   # First match in priority array wins

        }
    }
}

Write-Host "PRIMARY route: $primary"

Real Routing Example

Consider this hint:

"reverse engineer iOS app, need to bypass jailbreak detection using Frida"

1. Scoring phase:

   • R4 (iOS reverse) hits on ios/appstore patterns → 1 hit
   • R21 (Frida scripting) hits on frida pattern → 1 hit
   • R0 (generic) may hit on reverse engineer → 1 hit

2. Tie-break phase: All three have equal scores. The priority array shows R4 at position 0, R21 at position 10, R0 at position 40.

3. Result: PRIMARY = R4 (iOS reverse) due to priority ordering precedence.

Key Files Defining Priority Ordering

File	Role	Critical Lines
skills/config/routing.json	Source of truth for priority array and scoring rules	1-313
skills/MASTER-ROUTING.md	Human-readable priority table synchronized with JSON	52-96
skills/scripts/master-route.ps1	Production implementation of the routing algorithm	Full file
skills/scripts/verify-routing-coherence.ps1	CI validation ensuring array/table consistency	Full file

Summary

• Priority ordering resolves routing ties deterministically using author-controlled sequence
• Two-step process: keyword scoring creates candidates, priority array selects winner
• priority array position determines precedence—earlier IDs win ties
• fallbackId: "R0" catches unmatched hints with generic skill
• Consistency enforced by verify-routing-coherence.ps1 between JSON and Markdown documentation

Frequently Asked Questions

What happens if two routes have different scores but the lower-priority one scores higher?

The route with the higher score always wins, regardless of priority ordering. The priority array only activates when multiple routes share the identical highest hit count. Score takes absolute precedence over position in the priority list.

Can I modify the priority ordering without changing routing behavior for unambiguous hints?

Yes. Changing the priority array only affects tie-break situations where multiple routes receive equal top scores. Hints with a single clear winner or distinctly dominant scores route identically regardless of priority adjustments.

How does the system prevent priority drift between code and documentation?

The skills/scripts/verify-routing-coherence.ps1 script runs in CI to validate that routing.json's priority array matches the sequence defined in MASTER-ROUTING.md. Any divergence fails the build, ensuring the human-readable table always reflects actual runtime behavior.

Why does R0 appear last in the priority array rather than first?

R0 serves as the generic fallback for unmatched hints via fallbackId, not through priority ordering. Placing it last ensures that if it accidentally hits keywords simultaneously with specialized routes, those specific skills win the tie. This prevents over-generalization when precise tools are available.

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