popoffvg/context-research

Context Research

Procedure

Step 1: Check memory first

1. Use mcp__qmd__search with collection: "ctx" for the topic
2. Use mcp__qmd__deep_search with collection: "ctx" if few results
3. Use mcp__qmd__search with collection: "z-core" (Obsidian vault) as secondary fallback
4. Present what's already known

Step 2: New research (if memory insufficient)

1. Use mcp__firecrawl__firecrawl_search to find current information
2. Use mcp__firecrawl__firecrawl_scrape to read relevant pages
3. Synthesize findings into a clear summary

Configuration

Read insights_root from ~/.claude/memory-keeper.local.md YAML frontmatter. If the file is missing, stop and ask the user to create it with the required settings (see plugin README).

Step 3: Save results to memory

1. Always persist research results using the context-save skill procedure with type: common — research output is global/cross-project knowledge
2. Include sources (URLs, dates) in the Sources section
3. Update _summary.md and <insights_root>/INDEX.md

Output

Return both the answer AND confirmation of what was persisted to memory.

Autoresearch rules

Eval checklist:

1. Was persistent memory checked BEFORE performing web research?
2. Did web research only trigger when memory was insufficient (not by default)?
3. Were research results persisted to memory for future recall?
4. Did the agent present what was already known before adding new findings?

Test inputs:

• "Research Go context cancellation patterns" (likely in memory)
• "Research a brand new library released last week" (memory empty, needs web)
• "Look into why our CI is slow" (mixed: some memory, some new research)

Can change: memory check depth, web search strategy, persistence format, presentation order Cannot change: memory-first principle, QMD for memory, firecrawl for web, persist-results requirement Min sessions before eval: 5 Runs per experiment: 3