Unlimited skill

wuykjl — Tue, 16 Jun 2026 02:42:40 +0000

Unlimited skill engine — Hermes Agent 5-layer hybrid routing plugin.

Like a maître d' at a fine restaurant: you say "I'm hungry", they seat you at the best table — no need to flip through a 400-dish menu.

See the Difference

Me: "Help me optimize my website"

❌ Before: AI confused, spits out irrelevant text

✅ After: AI understands instantly, calls the "Performance Optimization" skill

Me: "Thanks, well said"

❌ Before: Still suggesting skills like a pushy salesperson

✅ After: A quiet "you're welcome" — just chatting normally

Architecture

代码
User Query
│
▼
[1] Task Gate (rules + LLM hybrid)
├── Not a task → skip (zero overhead)
└── Task →
│
▼
[2] Exact Trigger Match
├── Hit → Top-3 injected
└── Miss →
│
▼
[3] Trigger Word Overlap (overlap scoring)
├── Hit → Top-3 injected
└── Miss →
│
▼
[4] Levenshtein Fuzzy (typo tolerance)
├── Hit → Top-3 injected
└── Miss →
│
▼
[5] LLM Fallback (edge case catch-all)
→ Top-3 injected
Features

Unlimited skills — BM25 + trigger phrase matching, O(1) lookup, skill count doesn't affect latency
Bilingual CN/EN — 2,918 trigger phrases covering real Chinese and English user queries
Self-training loop — user corrections → feedback log → cron auto-training
Drift monitor — runs 366 test cases daily at 4AM, alerts if accuracy drops >5%
Zero noise — scores 0.5–1.0 have meaning; no BM25 random guesses
100% local — all indexes on-device, no external APIs
Install

代码
· bash
hermes skills install https://github.com/wuykjl/unlimited-skills
Install and go. Works in Chinese and English. Learns on its own. Monitors itself for drift.

Who's It For?

You say "write me a script", "deploy this", "check for security holes" — daily
300+ skills installed and you don't know which ones are actually being used
You don't want to waste time thinking about "which skill to pick"
Acknowledgments

This project's skill routing is built on the open skill ecosystems of:

ECC (Enterprise Coding Conventions) — 117-rule coding standard framework
Anthropic Skills — Official Agent Skills repository
Claude Code Plugins — Claude Code official plugin ecosystem
And all individual developers who contributed skills to the Hermes ecosystem

How I turned 18% skill hit rate into 95% — without calling an embedding API once

wuykjl — Sun, 14 Jun 2026 10:43:19 +0000

My Claude Code setup grew to 152 skills. Every session, all 152 descriptions got dumped into the system prompt. The LLM scanned them blindly. When I later checked, keyword matching hit the right skill 18% of the time.

I had three options:

Call text-embedding-3 on every query — costs money, adds latency
Accept 18% as "good enough"
Build something
What I built
neuro-skill — a hybrid skill router that runs entirely on my machine. No API calls. No GPU. 5ms per query.

It fuses five signals via Reciprocal Rank Fusion:

BM25 keyword recall → Cosine feature similarity → Graph spreading activation
→ Collaborative filtering personalization → Optional LLM rerank
The results
Core domains: 95% Hit@1, 100% Hit@3
Chinese queries: fixed from 0% to 87% with bilingual keyword coverage
Multi-skill orchestration: router.plan("review + fix + deploy") returns ordered execution steps
Personalization: learns which skills you pick, not just which ones rank highest
An independent tester validated it on 332 skills (Hermes + ECC) — v0.7.1 hit 77% overall.

Architecture decisions that mattered
Feature matrix, not embeddings. 50 hand-crafted features (17 broad domains + 32 precise languages/actions) beat TF-IDF by 5× without any model.
Graph over threshold. k-NN graph (k=adaptive) keeps diffusion working at any skill count — 0.5% density with full adjacency was dead on arrival.
RRF over weighted sum. Min-max normalization broke when BM25 scored 0–15 and cosine gave −1 to 1. Reciprocal Rank Fusion only cares about rank position.
LLM as opt-in, not default. Haiku rerank is a 5th signal you toggle on when you need semantic nuance. Otherwise it's zero-cost.
MCP from day one. Claude Code, Cursor, Codex, Windsurf — all auto-discover the tools. No plugin installs.
What I learned
The algorithm ceiling is real. BM25 + cosine + graph + RRF + CF + LLM rerank — that's six layers. After that, gains are marginal. The remaining bottleneck is feature coverage, not algorithm design.

Also: three independent projects (neuro-skill, agent-skill-finder, SkillRouter paper) converged on the same architecture from different starting points. That's when you know you're on the right track.

Links
GitHub: github.com/wuykjl/neuro-skill
pip install neuro-skill
MIT licensed. 54 tests. 28 commits over 12 hours.

I built this for myself. It turned out useful for others too. If your agent has 50+ skills and keyword matching is letting you down, try it.

DEV Community: wuykjl

Unlimited skill

How I turned 18% skill hit rate into 95% — without calling an embedding API once