Deep Context

Deep Context Skill

Cross-channel research agent that searches ALL connected systems for everything known about a topic, then synthesizes a comprehensive briefing. Searches beyond Atlassian to include Obsidian, Gmail, Slack, GitHub, and Granola.

When to Activate

• User says "deep context", "tell me everything about X", "research X"
• User says "what do we know about X?", "full context on X"
• User says "brief me on X", "context dump for X"
• Before writing a spec, RFC, or making a strategic decision
• When onboarding someone to a topic

How It Differs from Simple Search

Use /simple-search for quick factual questions. Use /deep-context for comprehensive research.

Input

Required:

• Topic: The subject to research (e.g., "ESP corridors", "MatchMaker redesign", "lead quality")

Optional:

• Time range: Defaults to "all time". User can specify "last 3 months", "Q1 2026", etc.
• Focus: "decisions", "status", "history", "people" - narrows the synthesis
• Output: "conversation" (default), "file", "confluence"

Execution

Step 0: Expand Search Terms

Use domain terminology expansion:

• Input "ESP" → also search "Explore Similar Programs", "multiplier", "loop yield"
• Input "search" → also search "Elasticsearch", "SERP", "relevancy", "RRF"
• Input "ads" → also search "Kevel", "Featured Listings", "sponsored"

For general topics, generate 2-3 synonym expansions.

Step 1: Parallel Data Gathering (Fan-Out)

Launch ALL searches in parallel using Agent tool or direct MCP calls:

Step 2: Deduplicate and Rank

• Same item across sources (e.g., Jira ticket mentioned in Slack) = merge, cite both
• Rank by: recency first, then richness of content
• Group by theme/subtopic if the topic is broad

Step 3: Synthesize Briefing

Build the structured output from merged results.

Output Format

# Deep Context: {Topic}

**Generated:** {timestamp}
**Sources searched:** {list of sources that returned results}
**Total items found:** {count across all sources}

---

## Executive Summary
{3-5 sentences: What is this topic, what's its current status, and what matters most right now}

## Timeline
| Date | Event | Source |
|:-----|:------|:-------|
| {date} | {What happened} | {Jira/Confluence/Slack/etc.} |
| {date} | {What happened} | {source} |
| {date} | {What happened} | {source} |

## Current Status
- **State:** {Active / Completed / Paused / Blocked}
- **Owner:** {Person or squad}
- **Last activity:** {date} in {source}
- {Key metrics or status indicators if available}

## Key Decisions
1. **{Decision}** ({date}): {What was decided and why}
   - Source: {link or reference}
2. **{Decision}** ({date}): {What was decided and why}
   - Source: {link or reference}

## Open Questions / Unresolved
- {Question or thread that hasn't been resolved}
- {Blocked item or pending decision}

## Key People
| Person | Role in This Topic | Recent Activity |
|:-------|:-------------------|:----------------|
| {name} | {their involvement} | {what they did recently} |

## All Sources
### Jira
- [{TICKET}] {title} - {status} ([link])

### Confluence
- {Page title} - last updated {date} ([link])

### Slack
- #{channel}: {thread summary} ({date})

### Obsidian
- {Note title} ({date}): {relevant excerpt}

### Gmail
- {Subject line} from {sender} ({date})

### GitHub
- {PR/Issue title} - {status} ([link])

---
*Generated by /deep-context*

Rules

• Exhaustive then concise: Search everything, but synthesize tightly. The briefing should be readable in 3-5 minutes.
• Source everything: Every claim must have a source citation.
• Conflict surfacing: If sources disagree, call it out explicitly (same pattern as domain-search).
• Recency weighting: More recent items get more prominence, but don't omit history.
• No speculation: If something is unclear, say "unclear" not "probably".
• Privacy: Don't include others' private Slack DMs. Only include threads where the user is a participant or in public channels.

Optimization

For very broad topics that return too many results:

1. Cap each source at 15-20 items
2. Prioritize last 90 days unless user specified otherwise
3. Group by subtopic and summarize groups rather than listing every item
4. Offer to "go deeper" on any subtopic

Fallback

If MCP tools are unavailable:

1. Always search Obsidian vault (local, always available)
2. Always check domain-search local context files
3. Note which sources were unreachable
4. Produce partial briefing with available data
5. Suggest: "Re-run when {source} is available for a complete picture"

Examples

User: "deep context ESP corridors"
→ Searches all sources for ESP, Explore Similar Programs, multiplier, corridor, loop yield
→ Produces full briefing with Q1 OKR data, Jira tickets, meeting decisions, Slack threads

User: "tell me everything about the MatchMaker redesign"
→ Searches all sources for MatchMaker, recommendation form, wizard, lead capture
→ Produces timeline from initial design to current implementation status

User: "deep context INS-1959"
→ Searches for the specific bug across Jira, Slack, GitHub PRs, meeting notes
→ Produces lifecycle view: reported → triaged → discussed → current state

User: "brief me on lead quality, focus on decisions, last 3 months"
→ Narrows time range, emphasizes Key Decisions section
→ Produces decision-focused briefing