popoffvg/context-find

Context Find

Core rule

Execute all search steps autonomously. Never ask the user for confirmation between steps. Run keyword search, semantic search, and fallback search without prompting. Only ask once — at the end — if nothing was found and you need the user to specify where else to look.

Usage

/context find <query>

Examples:

• /context find kubernetes debugging patterns
• /context find what do I know about the auth service
• /context find (shows all projects overview)

Search Procedure

1. 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).
2. If no query → read and display <insights_root>/INDEX.md
3. Use qmd_search with collection: "ctx" for keyword matching
4. If few results → use qmd_query with collection: "ctx" for semantic search (do this automatically, no confirmation needed)
5. If still insufficient → search qmd_search with collection: "z-core" (Obsidian vault) as fallback (do this automatically, no confirmation needed)
6. Use qmd_get or Read tool to retrieve full file content
7. If nothing found → ask the user once where to search (e.g. specific repo, web, docs URL) and use the appropriate tool
8. Present results with project/topic context and source file paths

List Procedure

1. Read <insights_root>/INDEX.md for project overview
2. For specific project → read <insights_root>/<project>/_summary.md
3. Present organized summary

Tool name mapping

| Use this | Not this | |---|---| | qmd_search | mcp__qmd__search | | qmd_query | mcp__qmd__deep_search / mcp__qmd__vector_search | | qmd_get | mcp__qmd__get |

Autoresearch rules

Eval checklist:

1. Did the search return results from persistent memory (not fabricated from training data)?
2. Were both keyword search and semantic search attempted when keyword results were sparse?
3. Did the agent present findings with source file paths (not just content)?
4. Was the insights_root config read before searching?
5. Did the agent complete the search without asking any yes/no confirmation questions mid-procedure?

Test inputs:

• "Recall what I know about Kubernetes debugging patterns"
• "Find my notes on the auth service architecture"
• "context find" with no query (should show INDEX.md overview)

Can change: search strategy, result presentation format, fallback search order, query expansion Cannot change: QMD tools as search backend, insights_root config requirement, memory-first principle Min sessions before eval: 5 Runs per experiment: 3