agentsclimarketplace

Knowledge architecture

Skill AVIDS2/avid-skill/skills/knowledge-architecture

Use when designing knowledge bases, RAG, wiki, knowledge graphs, memory for agents, or "how should we store docs for humans vs agents". Produces a clear consumer/form/retrieval decision. Avoids buzzword soup. Do NOT use for ordinary app CRUD databases.From its SKILL.md

Install
npx -y skills add AVIDS2/avid-skill --skill knowledge-architecture

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.5 KB, as published. Nobody here has run it

Knowledge Architecture

Separate three decisions people often mash together: who consumes, what form, how retrieval works.

Step 1 — Who consumes?

ConsumerOptimize for
HumansReadable structure, navigation, editability (Wiki / Markdown tree)
AgentsChunkability, metadata, stable IDs, tool-friendly retrieval
Both (default for products)Single source of truth humans edit + indexes agents query

Default bias: coexistence — humans write Markdown/Wiki; agents get hybrid retrieval over the same source. Avoid a second shadow knowledge base unless required.

Step 2 — What form?

FormWhen
Markdown / WikiMost products; docs, runbooks, specs
Structured DB recordsStrong schema entities (tickets, SKUs)
Knowledge graphRelationships are first-class (org charts, law cites, multi-hop entity links)
Vector-only corpusLarge unstructured dump with weak structure needs

MVP default: Markdown wiki (or /docs). Promote to graph when queries are relationship-heavy and hybrid RAG fails.

Step 3 — Retrieval pattern

PatternWhen
Full-text onlySmall corpus, precise keywords
Vector RAGSemantic questions over unstructured text
Hybrid (keyword + vector)Default production RAG
Agentic RAGMulti-step: plan → retrieve → critique → retrieve again
Graph RAGMulti-hop entity/relation questions

MVP default: hybrid retrieval. Use agentic RAG when single-shot retrieval quality is measurably insufficient.

Decision Output

## Consumers: human | agent | both
## Source form: markdown-wiki | db | graph | mixed
## Retrieval: fts | vector | hybrid | agentic | graph-rag
## Ingestion path
## Evaluation plan (5 real questions the system must answer)
## Non-goals

Anti-Patterns

  • "We need a knowledge graph" with no multi-hop questions
  • Separate Notion for humans + random vectors for agents with no sync
  • Agentic RAG theater on a 20-page corpus
  • Building custom RAG platform instead of using mature components (research-before-build)

Related

  • research-before-build, stack-defaults
  • references/knowledge-decision-tree.md

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.