| name | search |
| description | Search across all connected company tools in one unified query. Finds results from email, chat, documents, wikis, tickets, and other enterprise sources, then ranks and presents them in a single, actionable result set. TRIGGER when: user asks to search, find, look up, locate, or query for something across company tools, or asks "where is X", "find me X", "search for X", or uses a simple search query.
|
| argument-hint | <search query> [--source EMAIL|CHAT|DOCS|WIKI|TICKETS] [--from PERSON] [--since DATE] |
| user-invocable | true |
Enterprise Unified Search
Execute a search across all connected enterprise tools and return a merged, ranked, and deduplicated result set.
Search Execution Process
Step 1: Parse the Search Query
Extract structured parameters from the natural language query:
| Parameter | Extraction Pattern | Example |
|---|
| Keywords | Core topic words | "deployment process" |
| Source hint | "in email", "on Slack", "in Jira" | source:email |
| Person | "from Alice", "by the backend team" | from:alice@company.com |
| Date range | "last week", "in March", "since Q1" | since:2026-01-01 |
| Type | "document", "thread", "ticket", "page" | type:document |
| Project | "in Project Alpha", "for the migration" | project:alpha |
| Status | "open tickets", "resolved", "draft" | status:open |
Step 2: Execute Parallel Searches
Search all relevant sources simultaneously:
| Source | Query Adaptation |
|---|
| Email | Subject + body search; filter by sender, date, attachments |
| Chat | Message search; filter by channel, sender, reactions |
| Documents | Title + content search; filter by owner, type, folder |
| Wiki | Page title + content; filter by space, labels, author |
| Tickets | Summary + description + comments; filter by status, label |
| Calendar | Event title + description; filter by date, attendees |
| Code | File content + names + commit messages; filter by repo |
Step 3: Merge and Rank Results
Ranking factors by weight:
| Factor | Weight | Description |
|---|
| Text relevance | 35% | How closely content matches the query |
| Recency | 25% | More recent results rank higher |
| Engagement signals | 15% | Views, reactions, comments, shares |
| Source authority | 15% | Wiki > Docs > Email > Chat for factual queries |
| User affinity | 10% | Content from user's team, frequented channels, owned docs |
Step 4: Deduplicate
Remove or merge duplicate results:
- Same link shared across email, chat, and tickets — show once, note all sources
- Document with multiple versions — show latest, note version count
- Forwarded emails — show original thread, note forwards
Output Format
# Search Results: "[query]"
Found [N] results across [N] sources ([time] elapsed)
Filters applied: [list any active filters]
## Top Results
### 1. [Title or Subject]
**Source**: [Icon] [Source name] | **Date**: [date] | **By**: [author]
**Match**: [brief excerpt with **highlighted** query terms]
[Relevance: why this result ranked high]
### 2. [Title or Subject]
**Source**: [Icon] [Source name] | **Date**: [date] | **By**: [author]
**Match**: [brief excerpt with **highlighted** query terms]
### 3. [Title or Subject]
**Source**: [Icon] [Source name] | **Date**: [date] | **By**: [author]
**Match**: [brief excerpt with **highlighted** query terms]
[... up to 10 results ...]
## Results by Source
| Source | Results | Top Match |
|----------|---------|------------------------------------|
| Email | [N] | [title of best email result] |
| Chat | [N] | [title of best chat result] |
| Docs | [N] | [title of best doc result] |
| Wiki | [N] | [title of best wiki result] |
| Tickets | [N] | [title of best ticket result] |
## Suggested Refinements
- Add filter: [suggestion based on results, e.g., "from:engineering"]
- Narrow to: [specific source that had best results]
- Related searches: [alternative query suggestions]
Search Modifiers
Users can apply these modifiers to refine searches:
| Modifier | Syntax | Example |
|---|
| Exact phrase | "phrase" | "deployment pipeline" |
| Exclude term | -term | deployment -staging |
| Source filter | --source NAME | --source slack |
| Author filter | --from NAME | --from alice |
| Date filter | --since DATE | --since 2026-01-01 |
| Date range | --since DATE --until DATE | --since 2026-01-01 --until 2026-03-01 |
| Type filter | --type TYPE | --type document |
| Project filter | --project NAME | --project alpha |
| Sort order | `--sort relevance | date` |
| Result limit | --limit N | --limit 20 |
Source-Specific Search Tips
Email
- Search attachments by filename:
attachment:report.pdf
- Filter by thread length for substantive discussions:
replies:>3
- Look for decisions: combine with keywords like "approved", "decided", "go ahead"
Chat / Slack
- Search specific channels:
in:#channel-name
- Find shared links:
has:link
- Find files:
has:file
- Reactions as signals: messages with many reactions are often important
Documents
- Filter by document type: spreadsheet, presentation, document
- Search in shared drives or specific folders
- Recently modified documents often have the most current info
Wiki
- Search by space or category for scoped results
- Labels and tags are strong signals for relevance
- Parent page hierarchy indicates topic structure
Tickets
- Search by status: open, in-progress, resolved, closed
- Labels and components narrow results significantly
- Comments often contain the most useful context
Quality Standards
- Response time: Present initial results within 5 seconds; complete results within 15 seconds
- Minimum results: Always search at least 3 sources before reporting "no results"
- Context: Show enough context (50-100 chars around match) for users to evaluate relevance
- Freshness: Indicate index freshness if results may be stale
- Accuracy: Never fabricate results; only show actual matches from sources
Edge Cases
- Empty query: Ask the user what they are looking for; suggest recent activity or popular searches
- Very broad query (e.g., "project"): Ask for clarification or return results grouped by source with a suggestion to narrow
- No results: Report which sources were searched, suggest alternative terms, check for typos
- Single character or very short query: Require at least 2 meaningful characters; suggest expanding
- Special characters: Escape special characters in queries; warn if query syntax looks malformed
- Rate-limited source: Return available results and note which sources were partially searched
Follow-Up Actions
After presenting results, offer:
- "Want me to expand result #N with full content?"
- "Should I narrow these results by [suggested filter]?"
- "Want me to synthesize the top results into a summary?"
- "Should I search with different terms?"
Quality Checklist
