llm-wiki-choose

name	llm-wiki-choose
description	Help a user choose between ready-made LLM-Wiki tools, a local-first workflow, a team rollout, retrieval/GraphRAG systems, or a custom implementation. Use when the user asks what solution fits their needs, whether to use OpenWiki/Obsidian/qmd/GraphRAG, or whether to build their own.
license	MIT
compatibility	Designed for Agent Skills-compatible coding agents. Browse and cite fresh sources before making current tool recommendations.
metadata	{"author":"po4yka","version":"1.0.0","install_scope":"self-contained"}

LLM-Wiki Choose

Goal

Interview the user enough to recommend a practical path: adopt a ready-made solution, assemble a local-first workflow, roll out a team process, or build a custom LLM-Wiki.

When to use

• The user asks which LLM-Wiki approach fits: a ready-made tool, a local-first git workflow, a team rollout, a retrieval/GraphRAG system, or a custom build.
• The user explicitly asks whether OpenWiki, Obsidian, qmd, or a GraphRAG system suits their case.
• The user is comparing infrastructure options before committing to an architecture or before starting llm-wiki-paf-adoption.
• The user is still skeptical whether an LLM-Wiki is worth building at all.

Inputs

• User type (individual, team, company, product builder) and corpus type/scale.
• Privacy tier (public, internal, sensitive, regulated) and update/freshness needs.
• Preferred coding agent and existing tooling (Obsidian, GitHub, qmd, vector DB, GraphRAG, OpenWiki, custom scripts).
• Write tolerance (read-only advisory, PR-based writes, direct writes) and infrastructure/maintenance budget.

Procedure

1. Gather requirements

Ask only the questions that materially change the recommendation. Cover:

• user type: individual, team, company, product builder;
• corpus: code repo, research papers, docs folder, Obsidian vault, chat exports, web clips, PDFs, media transcripts;
• scale: rough number of sources/pages;
• update frequency and freshness needs;
• privacy: public, internal, sensitive, regulated;
• preferred agent: Claude Code, Codex, Cursor, OpenCode, GitHub Copilot, other;
• local-first and git requirements;
• write tolerance: read-only advisory, PR-based writes, direct writes;
• existing tools: Obsidian, GitHub, qmd, vector DB, GraphRAG, OpenWiki, custom scripts;
• budget for infrastructure and maintenance.

If the user already provided these facts, do not ask again.

2. Classify the case

Use this routing:

Case	Default recommendation
Small personal vault, local-first	Markdown + git + agent skills + rg; add qmd later.
Code repository docs	OpenWiki-style repo wiki or custom docs-as-code workflow.
Existing Obsidian vault	Refactor in place with protected human sections and lint gates.
Team/company knowledge	PR-based agent writes, CODEOWNERS, permissions, lint reports.
Large retrieval-heavy corpus	Hybrid search or graph/RAG system, with wiki as compiled surface if human review matters.
Product/plugin idea	Custom architecture with provenance, staged review and safe writes from day one.

3. Answer adoption objections before tool choice

If the user is still asking whether LLM-Wiki is worth it, invoke or summarize llm-wiki-faq before recommending tools.

Use the evidence hierarchy:

direct LLM-Wiki evidence -> adjacent RAG/memory/context evidence -> local pilot metrics

Do not claim that adjacent GraphRAG/memory evidence proves every LLM-Wiki workflow will work.

4. Decide whether Obsidian is needed

Obsidian is optional.

Recommend Obsidian when:

• the user already uses it;
• human reading/editing of Markdown is central;
• backlinks, graph view and local note UX matter;
• the workflow is personal or research-heavy.

Do not require Obsidian when:

• the target is repository documentation;
• the team already uses docs-as-code and pull requests;
• the user wants minimal CLI/git workflow;
• the user only needs agent-readable context.

5. Re-verify current external options

Before recommending a current tool, browse and cite:

• official docs or repository README;
• recent commits/releases/issues when maturity matters;
• license;
• local/cloud inference support;
• output format and data portability;
• provenance/review features;
• install path and supported agents.

6. Score options

Score 1-5 across:

• fit to use case;
• local-first/data ownership;
• setup effort;
• maintenance burden;
• provenance/trust controls;
• agent compatibility;
• scaling path;
• reversibility/lock-in.

7. Recommend one path

Give a primary recommendation and one fallback. Avoid listing every possible tool as equal.

Use this output:

## Recommendation

## Why this fits

## Evidence level

## Alternatives considered

| Option | Fit | Setup effort | Trust controls | Notes |
|---|---:|---:|---:|---|

## Setup plan

## Risks and review gates

## Next skill to run

Decision rules

• Prefer a thin local-first workflow before adding databases.
• Prefer ready-made tools for repo documentation and demos.
• Prefer custom implementation only when trust, provenance, scale, UX or integration needs justify it.
• Prefer PR-based writes for teams.
• Do not recommend current project maturity from memory; browse.
• For skeptical users, recommend a 20-50 source pilot and llm-wiki-eval rather than a big migration.

Output

A short markdown report following the section 7 template: Recommendation, Why this fits, Evidence level, an Alternatives considered table (Option / Fit / Setup effort / Trust controls / Notes), Setup plan, Risks and review gates, and Next skill to run.

Safety gates

• Stop before suggesting cloud ingestion for sensitive material unless the user explicitly accepts it.
• Do not recommend direct-write agents for team knowledge without review gates.
• Mark uncertain tool claims as verify-before-use.
• Do not make Obsidian mandatory unless the user's workflow actually benefits from it.