Deep Context

Deep Context

Comprehensive cross-channel research on any topic. Searches every connected tool in parallel, then synthesizes findings into a structured briefing. Use this when you need the full picture on a project, feature, incident, or initiative.

When to Activate

• User says "deep context on X", "tell me everything about X"
• User says "research X", "full context on X", "what do we know about X"
• User needs to get up to speed on an unfamiliar topic quickly

Input

Required:

• Topic: The subject to research (project name, feature, incident, concept, person)

Optional:

• Time range: How far back to search (default: 90 days)
• Depth: "quick" (summary only) or "deep" (full timeline + all sources, default)

Execution

Context Check

Before searching, read all four context files (context/company.md, context/product.md, context/competitors.md, context/personas.md) if they exist. Existing context informs search terms and avoids redundant research. After synthesis, offer to update context files with new knowledge discovered.

Step 1: Expand Search Terms

Before searching, generate 3-5 related search terms. Topics often have aliases, abbreviations, or related concepts.

Example: "authentication redesign" expands to: "auth redesign", "login", "SSO", "identity", "auth v2"

Step 2: Parallel Data Gathering (Fan-Out)

Search ALL connected MCP sources in parallel using each search term.

| Source | Method | What to Extract | |:-------|:-------|:----------------| | Jira | JQL text search | Epics, stories, bugs, comments mentioning topic | | Confluence | CQL text search | Specs, RFCs, design docs, meeting notes | | Slack | Message search across channels | Discussions, decisions, announcements | | GitHub | Code search, issue search, PR search | Implementation details, technical decisions | | Gmail | Thread search | Stakeholder communications | | Google Calendar | Event search | Meetings related to topic |

Step 3: Build Timeline

Order all findings chronologically. Identify:

• Origin: When did this topic first appear?
• Key milestones: Major decisions, launches, pivots
• Current state: Where things stand today
• Stakeholders: Who is involved and what role they play

Step 4: Synthesize

Merge findings into a coherent narrative. Resolve contradictions by preferring more recent sources.

Output Format

Output directly in conversation (do NOT create a file unless asked):

# Deep Context: {Topic}
**Generated:** {date} | **Sources searched:** {count} | **Time range:** {range}

## Summary
{3-5 sentence executive summary of the topic: what it is, current status, key stakeholders}

## Timeline
| Date | Event | Source |
|:-----|:------|:-------|
| {date} | {First mention or kickoff} | {source} |
| {date} | {Key decision or milestone} | {source} |
| {date} | {Most recent activity} | {source} |

## Current Status
- **State:** {active/stalled/completed/planned}
- **Owner:** {person or team}
- **Next milestone:** {what's coming}
- **Blockers:** {if any}

## Stakeholders
| Person | Role | Last Activity |
|:-------|:-----|:-------------|
| {name} | {role in this topic} | {what they last did} |

## Key Documents
- {Document title} ({source + link}): {1-line description}

## Open Questions
- {Unresolved question found in discussions}
- {Decision that appears pending}

## Sources
{List every source consulted, whether it returned results or not}

Rules

• Exhaustive search: Check every connected source. Do not skip any.
• Cite everything: Every claim links back to a specific source.
• Chronological timeline: Events ordered by date, not by source.
• Surface contradictions: If sources disagree, note it explicitly.
• No speculation: If something is unclear, list it under Open Questions.
• Respect scope: Stay focused on the topic. Related tangents go under a brief "Related Topics" note at the end.

Fallback

If MCP tools are unavailable:

1. Note which sources were unreachable.
2. Use web search as a supplement for public topics.
3. Produce the briefing with whatever was found, clearly marking gaps.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mshadmanrahman
• Source: mshadmanrahman/pm-pilot
• License: MIT
• Homepage: https://github.com/mshadmanrahman/pm-pilot#readme

Install and usage instructions live in the source repository linked above.