qmd

QMD for Coding Agents

FlowState-QMD is a local-first memory layer for coding agents. It turns markdown knowledge bases into project memory that is fast enough to feel built in.

What It Is Good At

• Repo memory: design docs, specs, ADRs, RFCs, changelogs
• Team memory: meeting notes, incident reviews, migration summaries
• Agent memory: anticipatory context before a reactive search step
• Trustworthy recall: inspectable snippets, doc IDs, and explain traces

Status

!qmd status 2>/dev/null || echo "Not installed: run 'qmd skill install --global' or 'npm install -g @tobilu/qmd'"

Start Here

1. Call fetch_anticipatory_context at the beginning of a coding turn when recent conversation context is available.
2. Use query when you need deeper retrieval or want to inspect the evidence directly.
3. Use get or multi_get to pull exact docs, ADRs, changelogs, or notes once you know what matters.

MCP Tool: fetch_anticipatory_context

Use this first when you want the agent to feel like it already knows the project context.

{
  "recent_conversation": "What changed in the auth rollback plan and why did we revert the migration?",
  "refresh": false,
  "lite_mode": false
}

Returns:

• cached anticipatory memories from FlowState when available
• or a fresh live query over the current project memory

MCP Tool: query

{
  "searches": [
    { "type": "lex", "query": "\"auth rollback\" migration changelog" },
    { "type": "vec", "query": "why did we revert the authentication migration?" }
  ],
  "collections": ["docs", "notes"],
  "intent": "coding agent project memory for a migration debugging task",
  "limit": 8
}

Query Types

Type	Method	Best for
lex	BM25	exact names, code terms, filenames, changelog phrases
vec	Vector	natural-language engineering questions
hyde	Vector	nuanced answers when you know what the result should sound like

Writing Better Coding-Agent Queries