Skill Retriever semantic skill discovery for AI agents via 10K-category taxonomy

AgentSkillOS released Skill Retriever, a semantic skill discovery plugin for Hermes AI agents that uses a 10,000-category taxonomy to surface relevant skills from a corpus of 1,200+ skills. The plugin runs as a pre_llm_call hook with zero core modification and no additional API cost, transforming skill discovery from a flat catalog into a search engine. It pre-filters skills to the top-5 most relevant per query, addressing the limitations of pure semantic retrieval by navigating a capability hierarchy to find functionally relevant skills.

AgentSkillOS-powered semantic skill retrieval for Hermes Agent. Pre-filters 1,200+ skills 998 community corpus + 211 Hermes skills organized in a 10,000-category capability taxonomy to the top-5 most relevant per query. Runs as a Hermes pre llm call plugin — zero core modification, zero additional API cost borrows your existing Hermes LLMs via borrow-mode . Pure semantic retrieval prioritizes textual similarity and misses skills that look unrelated in embedding space but are crucial for solving the task. Our LLM + Skill Tree navigates the capability hierarchy to surface non-obvious but functionally relevant skills. Left: Pure semantic retrieval is narrow and myopic. Right: Skill Tree navigation surfaces functionally relevant skills the embedding space hides. Skills are organized into a coarse-to-fine capability hierarchy. At scale, this is the difference between finding the right skill and drowning in an invisible pile. The 10,000-category capability tree — the structure our 1,200 skills are mapped into. User Query │ ▼ ┌──────────────────────────────────────┐ │ pre llm call hook plugin │ │ Checks DISABLE flag, skips short Qs │ └──────────────┬───────────────────────┘ │ ▼ ┌──────────────────────────────────────┐ │ Searcher.search │ │ 1. Load capability tree from YAML │ │ 2. LLM-navigate tree select nodes │ │ 3. Parallel child search ThreadPool │ │ 4. LLM prune dedup + rank │ └──────────────┬───────────────────────┘ │ ▼ ┌──────────────────────────────────────┐ │ Hint Injection │ │ Prepends top-5 skill hints as │ │ natural-language block. LLM may call │ │ skill view name to load any. │ └──────────────────────────────────────┘ Hermes already ships with skill discovery — every user-installed skill appears in the <available skills block of the system prompt. The LLM scans this flat list every turn and calls skill view when needed. For small sets it works fine. skill-retriever adds a semantic retrieval layer that transforms skill discovery from "read the catalog" into "search for what you need": | Dimension | Hermes OOTB | skill-retriever | |---|---|---| Skill source | Your local ~/.hermes/skills/ only ~100-200 | Community corpus 998 + Hermes skills 200 = 1,198 total | Discovery | Flat name+desc list in system prompt every turn | LLM-navigated taxonomy tree → top-5 relevant injected as hints | Token cost | Every turn burns tokens for all skills, even irrelevant ones | Zero system prompt overhead — hints only in user message, only when found | Categorization | Filesystem directory names | 10,000-category AgentSkillOS capability taxonomy | Scales to | ~200 skills before prompt bloat | 10K+ tree handles it | Latency per turn | 0 passive — always visible | +1-3 cheap LLM calls for tree traversal when it has results | Community corpus | No | Yes — 998 community skills alongside yours | The difference: OOTB gives you a flat skill catalog you read every turn. skill-retriever turns it into a search engine — describe what you need, the tree navigates to the right category, and only relevant suggestions appear. The tradeoff is a small latency cost per turn vs constant system prompt bloat. git clone https://github.com/ChonSong/skill-retriever.git cd skill-retriever bash scripts/install.sh hermes gateway restart Every skill carries a source tag and a safety scan result : | Badge | Meaning | |---|---| 🔒hermes | Installed via Hermes — trusted | 🌐community | From AgentSkillOS corpus — unreviewed | ⚠️ suffix | Flagged by safety scan — review before loading | All 1,200 skills were scanned for dangerous patterns rm -rf / , curl | sh to raw IPs, base64 payloads, crypto miners . Zero flagged — every match was standard installer documentation inside code blocks. python -m skill retriever search "set up CI/CD pipeline" python -m skill retriever build rebuild capability tree python -m skill retriever list list all skills in corpus python -m skill retriever info system info + tree stats All settings via environment variables — no config files needed. | Env Variable | Default | Description | |---|---|---| SKILL RETRIEVER DISABLE | — | Set 1 to disable entirely | SKILL RETRIEVER LLM MODEL | gpt-4o | LLM model for skill gate | SKILL RETRIEVER LLM API KEY | OPENAI API KEY | API key | SKILL RETRIEVER LLM BASE URL | OPENAI BASE URL | Base URL | SKILL RETRIEVER BRANCHING FACTOR | 3 | Tree branching search | SKILL RETRIEVER MAX PARALLEL | 5 | Parallel search branches | SKILL RETRIEVER TEMPERATURE | 0.3 | LLM temperature | SKILL RETRIEVER PRUNE | true | Enable dedup/ranking step | SKILL RETRIEVER TREE PATH | bundled tree 10000.yaml | Override capability tree | See ARCHITECTURE.md /ChonSong/skill-retriever/blob/master/ARCHITECTURE.md for a technical deep-dive covering: - Capability tree structure and build process - LLM node selection algorithm - Searcher internals parallel search, early stop, pruning - Plugin hook integration - Directory layout - Hermes Agent v0.18+ - Python 3.10+ - ~500MB for capability tree index - ~4GB for full skill corpus optional, for rebuilding tree skill-retriever/ ├── plugin/ Hermes plugin pre llm call hook ├── src/ │ ├── skill retriever/ Core engine │ │ ├── cli.py CLI search, build, list, info │ │ ├── search/ Searcher multi-level LLM tree search │ │ ├── tree/ Tree builder, schema, prompts, scanner │ │ └── capability tree/ Pre-built trees YAML + HTML │ └── scanner.py Hermes skills scanner ├── data/ Skill corpus gitignored ├── tests/ 40 tests ├── scripts/install.sh One-click Hermes plugin install └── ARCHITECTURE.md MIT. Built on AgentSkillOS https://github.com/ynulihao/AgentSkillOS MIT .