research

name	research
description	Search QMD-indexed project and main cross-project LLM wikis for past patterns, decisions, solutions, and pitfalls before planning or implementation. Use before feature planning, bug fixing, refactoring, architecture work, or code changes when a project wiki may exist.

Wiki Research

Search the project wiki and main cross-project wiki before planning or implementation. Produce a concise Past Knowledge section that downstream planning or coding can use.

Step 1: Identify Context

Determine:

• Current project name: basename "$(git rev-parse --show-toplevel 2>/dev/null || pwd)"
• Parent of project: dirname "$(git rev-parse --show-toplevel)"
• Whether wiki/ exists in the repo.
• Whether .llm-wiki/config.json exists and contains main_wiki_path.
• Whether ~/wikis/master/wiki/ exists.
• Whether ~/wikis/main/wiki/ exists.
• Whether <parent-of-project>/wikis/master/wiki/ exists.
• Whether <parent-of-project>/wikis/main/wiki/ exists.
• Whether QMD MCP tools, qmd CLI, or only rg are available.

If no main cross-project wiki exists, say that explicitly. Ask the user for a main wiki folder when cross-project context is important, or suggest running bootstrap on Claude Code/Codex or wiki-bootstrap on Pi/OpenClaw to create <parent-of-project>/wikis/master/wiki/.

Step 2: Search With the Best Available Tool

Preferred: QMD MCP

Use at most two QMD MCP query calls. Do not fire many QMD calls in parallel; combine related searches into the searches array.

Project search:

{
  "searches": [
    { "type": "lex", "query": "<3-5 key terms>" },
    { "type": "vec", "query": "<natural-language task question>" }
  ],
  "collections": ["<current-project-name>"],
  "limit": 10
}

Main cross-project search:

{
  "searches": [
    { "type": "lex", "query": "<key terms>" },
    { "type": "vec", "query": "patterns for <feature domain>" }
  ],
  "collections": ["master"],
  "limit": 10
}

Use the master QMD collection for the main cross-project wiki when available. If the environment uses a main collection instead, query main.

Read the top 3-5 relevant pages with QMD get or multi-get.

Fallback: QMD CLI

qmd query "<topic>" --collection "<current-project-name>"
qmd query "<topic>" --collection master
# or, if the cross-project wiki is indexed as main:
qmd query "<topic>" --collection main

Fallback: ripgrep

rg "<key terms>" wiki/ --type md -C 2
rg "<key terms>" "<main_wiki_path from .llm-wiki/config.json>" --type md -C 2
rg "<key terms>" ~/wikis/master/wiki/ --type md -C 2
rg "<key terms>" ~/wikis/main/wiki/ --type md -C 2
rg "<key terms>" ../wikis/master/wiki/ --type md -C 2
rg "<key terms>" ../wikis/main/wiki/ --type md -C 2

Skip missing directories gracefully.

Step 3: Read Important Pages

Prioritize:

• wiki/decisions.md
• wiki/technical-debt.md
• wiki/gaps.md
• wiki/architecture.md
• <main_wiki_path from .llm-wiki/config.json>/patterns.md
• <main_wiki_path from .llm-wiki/config.json>/learnings.md
• ~/wikis/master/wiki/patterns.md
• ~/wikis/master/wiki/learnings.md
• ~/wikis/main/wiki/patterns.md
• ~/wikis/main/wiki/learnings.md
• <parent-of-project>/wikis/master/wiki/patterns.md
• <parent-of-project>/wikis/master/wiki/learnings.md
• <parent-of-project>/wikis/main/wiki/patterns.md
• <parent-of-project>/wikis/main/wiki/learnings.md

Only cite or summarize pages actually read.

Step 4: Produce Past Knowledge

Return this structure:

### Past Knowledge

**Relevant wiki pages:**
- ...

**Applicable patterns:**
- ...

**Past decisions that constrain this work:**
- ...

**Known pitfalls:**
- ...

**Reusable components:**
- ...

**Gaps this work could fill:**
- ...

If no wiki exists, say that explicitly and suggest running bootstrap on Claude Code/Codex or wiki-bootstrap on Pi/OpenClaw.

Rules

• Search wiki context before codebase exploration when wiki context exists.
• Do not claim wiki facts unless a page was read.
• Prefer concise synthesis over dumping search results.
• Preserve uncertainty and missing coverage as gaps.