| name | search-strategy |
| description | Plan and execute a multi-source search strategy for complex queries that require searching across multiple enterprise tools with different query syntaxes, filters, and relevance models. Decomposes broad questions into targeted sub-queries and merges results into a ranked answer set. TRIGGER when: user asks to plan a search, needs help finding something complex, wants a search strategy, asks "how should I search for X", or has a multi-faceted research question.
|
| argument-hint | <complex query or research question> [--sources SOURCE1,SOURCE2] |
| user-invocable | true |
Multi-Source Search Strategy
Plan, execute, and refine search strategies for complex enterprise queries that span multiple tools and require iterative refinement.
Strategy Planning Process
Step 1: Analyze the Query
Break down the user's question into searchable components:
| Component | Description | Example |
|---|
| Core concepts | Primary topics or entities | "authentication", "SSO" |
| Qualifiers | Narrowing attributes | "for mobile app", "in Q3" |
| Entity types | People, projects, teams, dates | "by security team", "Project X" |
| Intent | What the user actually needs | Decision history, how-to, status |
| Time constraints | Recency requirements | "last month", "current" |
Step 2: Select Target Sources
Evaluate which sources are most likely to contain relevant results:
| Source Type | Best For | Query Style |
|---|
| Email | Decisions, approvals, external communication | People + topic |
| Chat / Slack | Quick decisions, links, informal context | Channel + keywords |
| Documents | Specs, proposals, formal write-ups | Title + content |
| Wiki | Processes, architecture, onboarding, policies | Topic + category |
| Tickets / Issues | Implementation details, bugs, requirements | Labels + text |
| Code | Technical implementation, comments, configs | Symbol + file path |
| Calendar | Meeting context, attendees, scheduling | People + date range |
Source selection matrix:
| Query Intent | Primary Sources | Secondary Sources |
|---|
| "How does X work?" | Wiki, Docs, Code | Tickets, Chat |
| "What was decided?" | Email, Chat, Docs | Tickets, Wiki |
| "Who is responsible?" | Tickets, Wiki, Chat | Email, Calendar |
| "What is the status?" | Tickets, Chat, Email | Docs, Calendar |
| "When did X happen?" | Email, Chat, Tickets | Docs, Calendar |
| "Why was X done?" | Email, Chat, Docs | Tickets, Code |
Step 3: Formulate Sub-Queries
Transform the original query into source-specific sub-queries:
Original: "Why did we switch from Redis to DynamoDB for session storage?"
Sub-queries:
1. Wiki → "session storage" OR "DynamoDB" OR "Redis migration"
2. Docs → "session" AND ("DynamoDB" OR "Redis") — type:proposal OR type:design-doc
3. Email → "session storage" OR "Redis replacement" — from:engineering-leads
4. Chat → "DynamoDB sessions" OR "Redis sessions" — channels:#backend,#architecture
5. Tickets → labels:infrastructure "session" AND ("Redis" OR "DynamoDB")
6. Code → path:*session* OR path:*config* "dynamodb" — recent commits
Step 4: Define Execution Plan
# Search Execution Plan
## Query: [original query]
## Decomposition:
- Concept A: [term variations and synonyms]
- Concept B: [term variations and synonyms]
- Filter: [time range, people, teams]
## Execution Order:
1. [High-priority source] — [specific query] — Expected: [what we hope to find]
2. [Medium-priority source] — [specific query] — Expected: [what we hope to find]
3. [Lower-priority source] — [specific query] — Expected: [what we hope to find]
## Refinement Triggers:
- If Step 1 finds [X], narrow Step 2 to [Y]
- If Step 1 finds nothing, broaden Step 2 to [Z]
- If conflicting results, add Step 4: [verification query]
Step 5: Execute and Merge Results
Result Ranking Framework
Score each result on four dimensions:
| Dimension | Weight | Criteria |
|---|
| Relevance | 40% | How directly it answers the query |
| Authority | 25% | Source reliability, author expertise, formality level |
| Recency | 20% | How current the information is |
| Uniqueness | 15% | Does it add information not found in other results |
Composite score: (Relevance * 0.4) + (Authority * 0.25) + (Recency * 0.2) + (Uniqueness * 0.15)
Output Format
# Search Strategy Report
Query: [original question]
Sources searched: [N] | Results found: [N] | Top results: [N]
## Search Plan Executed
| Step | Source | Query Used | Results |
|------|----------|-------------------------------|---------|
| 1 | [Source] | [query] | [N] |
| 2 | [Source] | [query] | [N] |
## Top Results (Ranked)
### 1. [Result Title] — [Source] — Score: [X/10]
- **Relevance**: [why this matches]
- **Key excerpt**: "[relevant quote]"
- **Date**: [date] | **Author**: [author]
### 2. [Result Title] — [Source] — Score: [X/10]
- **Relevance**: [why this matches]
- **Key excerpt**: "[relevant quote]"
- **Date**: [date] | **Author**: [author]
## Answer Summary
[Synthesized answer based on top results]
## Search Refinement Suggestions
- To find more: try [broader query suggestion]
- To narrow down: add [filter suggestion]
- Unexplored sources: [sources not yet queried and why they might help]
Query Optimization Techniques
| Technique | When to Use | Example |
|---|
| Synonym expansion | Initial search yields few results | "auth" → "authentication", "login" |
| Phrase matching | Too many irrelevant results | "session storage" (exact phrase) |
| Author filtering | Known domain experts exist | from:jane.doe OR from:john.smith |
| Date narrowing | Topic changed over time | after:2025-01-01 |
| Negative filtering | Known false positives | NOT "session recording" |
| Channel scoping | Known relevant channels | in:#backend-eng |
| Label/tag filtering | Structured metadata available | label:architecture, tag:approved |
Iterative Refinement
After initial results, determine if refinement is needed:
- Too many results (>50): Add filters, use phrase matching, narrow date range
- Too few results (<3): Expand synonyms, broaden date range, add sources
- Wrong results: Analyze why results are off-topic, add negative filters
- Partial answer: Identify the gap and formulate a targeted follow-up query
Edge Cases
- Ambiguous queries: Ask the user to clarify before executing; present 2-3 interpretations
- No results anywhere: Suggest the information may not exist in connected sources; recommend who to ask
- Cross-language content: Note if sources contain content in multiple languages; expand queries accordingly
- Acronyms and jargon: Expand acronyms in queries and search for both forms
- Stale indexes: If a source's index is known to be delayed, warn about potential missing recent items
Quality Checklist
