Receber por email
← Voltar para o catálogo

qmd

Search the vault using QMD semantic search. Use PROACTIVELY before reading files directly — whenever the user asks about past decisions, incidents, people, meetings, architecture, patterns, or any vault content. Always prefer QMD over Grep/Glob for vault queries. Also use after creating/editing notes to check for duplicates and related content.

12estrelas
Atualizado há 6 meses

Ver no GitHub ↗ Licença: MIT

Como adicionar

/plugin marketplace add superuser-pal/awesome-second-brain

O comando exato pode variar conforme o repositório. Confira o README no GitHub.

Para o autor da skill

Cole no README do seu repo

Mostra que sua skill está catalogada na Skillteca, gera backlink e tráfego rastreável.

[![Skillteca](https://www.skillteca.com.br/api/badge/qmd-superuser-pal/svg)](https://www.skillteca.com.br/skills/qmd-superuser-pal)
#ai

Skills relacionadas

Ver todas de Produtividade →

Alerta por categoria

Receba novas skills de Produtividade toda segunda

1 email curto, só com as skills novas de Produtividade. 4 minutos de leitura, sem spam, cancela com 1 clique.

Você confirma o email no primeiro envio. Sem spam. Cancela com 1 clique.

Ler descrição completa

QMD — Vault Semantic Search

Before reading vault files directly, search with QMD first. It returns relevant snippets without burning context on full file reads.

Workflow Routing

| Workflow | Trigger | File | |----------|---------|------| | ask | When user runs /ask or asks a complex question needing direct answers | ../commands/qmd/ask.md | | search | When user runs /search or asks for a list of results | ../commands/qmd/search.md | | context | When user runs /context or asks for all context on a topic | ../commands/qmd/context.md | | reindex | When user runs /reindex or needs to refresh the QMD index manually | ../commands/qmd/reindex.md |

Commands

Search (pick one per query)

  • qmd query "..." --json -n 10 — Best quality. Hybrid BM25 + vector + LLM reranking. Use for complex or conceptual queries.
  • qmd search "..." --json -n 10 — Fast BM25 keyword. Use for exact terms, names, ticket numbers, dates.
  • qmd vsearch "..." --json -n 5 — Semantic only. Use for exploratory queries where you don't know the exact words.

Retrieve

  • qmd get "path/to/file.md" — Full document by path.
  • qmd get "#docid" — Full document by ID (from search results).
  • qmd multi-get "work/06_ORG/PEOPLE.md" --json — Batch retrieve people context.

Index Management

  • qmd update — Re-index after file changes (fast, ~1-2s incremental).
  • qmd embed — Regenerate vector embeddings (slower, run after bulk changes).

When to Search

  • User mentions a past decision, incident, person, project → qmd query
  • User asks "what did we decide about X" → qmd query
  • User mentions a person by name → qmd search "<name>"
  • Before creating a new note → qmd vsearch "<topic>" to check for existing content
  • After creating a note → qmd vsearch "<note title>" to find notes that should link to it
  • Loading context for review prep → qmd get "work/05_REVIEW/EVIDENCE.md"
  • Loading 1-on-1 context → qmd search "<person name> 1-1"

After Bulk Changes

Run qmd update && qmd embed to keep the index fresh. The SessionStart hook does qmd update automatically, but qmd embed should be run explicitly after sessions that create many notes.

Compartilhar X LinkedIn