Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add po4yka/llm-wiki-skills --skill llm-wiki-retrieval-architectgit clone --depth 1 https://github.com/po4yka/llm-wiki-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/po4yka/llm-wiki-skills/llm-wiki-retrieval-architect)<a href="https://agentmods.dev/skills/po4yka/llm-wiki-skills/llm-wiki-retrieval-architect"><img src="https://agentmods.dev/badge/skills/po4yka/llm-wiki-skills/llm-wiki-retrieval-architect/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/po4yka/llm-wiki-skills/llm-wiki-retrieval-architect"><img src="https://agentmods.dev/badge/skills/po4yka/llm-wiki-skills/llm-wiki-retrieval-architect.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00082 | $0.02412 |
| Opus 5 | $0.00041 | $0.01206 |
| Sonnet 5 | $0.00016 | $0.00482 |
| Haiku 4.5 | $0.00008 | $0.00241 |
Grade A, and why
llm-wiki-retrieval-architect scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 263 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LLM-Wiki Retrieval Architect
Goal
Choose the smallest retrieval tier that solves the user's measured retrieval failure while preserving inspectable Markdown, provenance, access filters and rebuildable indexes.
Use references/docs/16-retrieval-architecture.md as the reference playbook for detailed technology comparisons, metadata schema, GraphRAG lane design, evaluation gates, security controls and source URLs to re-check.
When to use
- The wiki's search misses relevant pages even though the content exists (lexical miss, stale index, poor ranking, thin context, or multi-hop miss).
- You need to pick or size a retrieval tier (grep-only, lexical index, hybrid semantic search, reranker, parent/context retrieval, graph-aware retrieval, or product storage) for a Markdown/wiki corpus.
- You must define index contracts, metadata/access filters, chunking policy, or graph edge schema for a new or growing wiki.
- Retrieval needs to respect tenant, sensitivity, or review-state boundaries before rollout.
- Whole-stack local-first tooling choices belong in
llm-wiki-local-first-stack; use this skill once retrieval specifically is the bottleneck.
Inputs
- Corpus size and source types.
- Current folder structure and wiki page schema.
- Query examples that failed.
- Privacy/local-first requirements.
- Latency, cost and deployment constraints.
- Existing tools:
rg, SQLite, vector DB, GraphRAG, LlamaIndex, Haystack, MCP. - Team/tenant/security requirements.
- Evaluation set or recent real questions, if available.
Procedure
1. Diagnose the retrieval failure
Classify the failure:
| Failure | Meaning | Typical fix |
|---|---|---|
| lexical miss | Keywords differ but concept is the same. | Add dense retrieval or query expansion. |
| exact-token miss | IDs, filenames, names or code symbols are missed. | Improve lexical/FTS lane; do not rely on dense-only search. |
| stale index | Search misses newly added or changed pages. | Fix rebuild/watch/CI triggers. |
| poor metadata | Results cannot filter by source, status, sensitivity or date. | Add required metadata fields and query-time filters. |
| poor ranking | Relevant result exists in top 20-100 but not top 5. | Add reranker. |
| thin context | Retrieved chunk is relevant but too small to answer. | Add parent-child or contextual retrieval. |
| synthesis miss | Individual pages are found but no cross-page answer exists. | Add synthesis pages or graph lane. |
| multi-hop miss | Relationships across entities/concepts are not traversed. | Add graph-aware retrieval or GraphRAG lane. |
| provenance miss | The result is plausible but unsupported by raw sources. | Add citation metadata and claim-support checks. |
| permission miss | Retrieval can expose restricted content. | Stop and fix filters/security before tuning relevance. |
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 263 lines · 82 tokens per session scan A 6e14a433e72b
llm-wiki-retrieval-architect is a skill published in the GitHub repository po4yka/llm-wiki-skills (3 stars, last pushed 19d ago), licensed MIT. It adds 82 tokens to every session and 2,412 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
knowledge-base-interop
Two-layer knowledge architecture — curate in LLM Wiki, search in Astra KB. Covers input classification, source-to-wiki ingestion, wiki-to-KB export (batch + realtime), semantic chunking, SAG extraction, health checks, and multi-format export.
azure-search-documents-dotnet
Azure AI Search SDK for .NET (Azure.Search.Documents). Use for building search applications with full-text, vector, semantic, and hybrid search. Covers SearchClient (queries, document CRUD), SearchIndexClient (index management), and SearchIndexerClient (indexers, skillsets). Triggers: "Azure Search .NET"…
azure-search-documents-ts
Build search applications using Azure AI Search SDK for JavaScript (@azure/search-documents). Use when creating/managing indexes, implementing vector/hybrid search, semantic ranking, or building agentic retrieval with knowledge bases.
convex-agent
Add an AI agent / RAG backend (@convex-dev/agent) to the Convex app.
wegent-knowledge
Knowledge base management and search tools for Wegent. Provides capabilities to list, create, update, and search knowledge bases and documents using RAG retrieval. Use this skill when the user wants to manage knowledge bases, documents, or search for information programmatically.
langchain4j-spring-boot-integration
Provides integration patterns for LangChain4j with Spring Boot. Configures AI model beans, sets up chat memory with Spring context, integrates RAG pipelines with Spring Data, and handles auto-configuration, dependency injection, and Spring ecosystem integration. Use when embedding LangChain4j into Spring Boot…