Coding Agents15.2ms Reflex$0 / MIT License
16,900 active downloads
SkillRanker
Benchmark ranking classifier for dynamic multi-agent skill dispatch
jevproxy // skillranker (reflex kernel)
$npx jevproxy run skillranker
JevProxy Intercept Resolved(Mode: DIRECT_REFLEX)
15.2ms|Cost: $0.0001|0 reasoning tokens burned
STDOUT • Tool Call Output:Status: 200 OK
agent.dispatch(skillranker_dispatch)
payload: { "query": { "type": "string", "description": "User input prompt" }, "candidate_skills": { ...
✔ Decision short-circuited in 15.2ms without roundtrip to frontier LLM.
Upstream token bill saved: $0.0165 on this turn.
SkillRanker
Compatible with Cursor, Claude Code, Windsurf, OpenCode
TRADITIONAL LLM CALL:UNOPTIMIZED
• Median Latency: 1540 ms
• Cost per Turn: $0.0165
• Mode: Full KV Cache Reload & TTFT Prefill
JEVPROXY REFLEX KERNEL:77x FASTER
• Median Latency: 15.2 ms
• Cost per Turn: $0.0001
• Accuracy: 96.4% deterministic
1-Click CLI Execution
Run this tool accelerated through the JevProxy gateway without manual wiring:
npx jevproxy run skillrankerOpenAI / Anthropic Tool Schema
JSON SpecificationPaste this schema into your agent tools definition or Cursor extensions:
{
"type": "function",
"function": {
"name": "skillranker_dispatch",
"description": "Score candidate skills against current turn state and pick top skill",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "User input prompt"
},
"candidate_skills": {
"type": "array",
"items": {
"type": "string"
},
"description": "Available skill identifiers"
}
},
"required": [
"query",
"candidate_skills"
]
}
}
}Technical Architecture & Usage
Tool Selection Scalability
As agent frameworks add dozens of MCP tools and custom skills, prompt sizes swell to 20,000+ tokens just for tool schemas. SkillRanker prunes tool candidates down to the relevant 3 in 15.2ms.
Accelerate SkillRanker with JevProxy
Get 5,000,000 free decision tokens. Eliminate the 3-second tool freeze in Cursor and Claude Code in under 60 seconds.