Document Intelligence
This skill supports three modes: Create, Update, and Find.
Mode Detection
| Signal | Mode | Confidence |
|---|---|---|
| "update", "add items", "finalize" in input | UPDATE | 100% |
| File path provided | UPDATE | 100% |
| "create", "new", "conduct" in input | CREATE | 100% |
| "find", "search", "list" | FIND | 100% |
| "the retrospective", "our retro" | UPDATE | 85% |
| Just project/quarter | CREATE | 60% |
Threshold: ≥85% auto-proceed | 70-84% state assumption | <70% ask user
Mode Behaviors
CREATE: Generate complete new retrospective using template below.
UPDATE:
- •Check document registry first, then search user's structure
- •Preserve participant input and prior discussions
- •Update actions, learnings, or follow-up items
- •Show diff summary
FIND: Check registry, then search user's folders for retrospectives.
Conduct a Structured Retrospective for the specified project, launch, or time period.
V2V Phase
Phase 6: Learning & Adaptation Loop - This skill captures learnings for continuous improvement.
Output Structure
1. Retrospective Overview
- •Scope: What we're reviewing
- •Timeframe: Period covered
- •Participants: Who contributed
- •Facilitator: Who led
2. Context Summary
- •Original goals/objectives
- •Key deliverables planned
- •Resources allocated
- •Timeline planned vs actual
3. What Went Well 🟢
| Success | Impact | Why It Worked | How to Replicate |
|---|---|---|---|
| [Success 1] | High/Med/Low | [Analysis] | [Action] |
| [Success 2] | High/Med/Low | [Analysis] | [Action] |
| [Success 3] | High/Med/Low | [Analysis] | [Action] |
4. What Didn't Go Well 🔴
| Challenge | Impact | Root Cause | Prevention |
|---|---|---|---|
| [Challenge 1] | High/Med/Low | [Cause] | [Action] |
| [Challenge 2] | High/Med/Low | [Cause] | [Action] |
| [Challenge 3] | High/Med/Low | [Cause] | [Action] |
5. What We Learned 💡
| Learning | Evidence | Application |
|---|---|---|
| [Learning 1] | [What showed us this] | [How to apply] |
| [Learning 2] | [What showed us this] | [How to apply] |
| [Learning 3] | [What showed us this] | [How to apply] |
6. Action Items
| Action | Owner | Due Date | Priority |
|---|---|---|---|
| [Action 1] | [Owner] | [Date] | P1/P2/P3 |
| [Action 2] | [Owner] | [Date] | P1/P2/P3 |
| [Action 3] | [Owner] | [Date] | P1/P2/P3 |
7. Process Improvements
| Process | Current State | Desired State | How to Improve |
|---|---|---|---|
| [Process 1] | [Current] | [Desired] | [Improvement] |
| [Process 2] | [Current] | [Desired] | [Improvement] |
8. Decisions to Revisit
| Decision | Original Rationale | New Evidence | Recommendation |
|---|---|---|---|
| [Decision 1] | [Why we decided] | [What we learned] | Keep/Modify/Reverse |
| [Decision 2] | [Why we decided] | [What we learned] | Keep/Modify/Reverse |
9. Recognition
- •People who went above and beyond
- •Teams that collaborated exceptionally
- •Contributions worth highlighting
10. Follow-up Plan
| Item | Owner | Check-in Date | Success Criteria |
|---|---|---|---|
| [Item 1] | [Owner] | [Date] | [Criteria] |
| [Item 2] | [Owner] | [Date] | [Criteria] |
Retrospective Facilitation Notes
For facilitators:
- •Create psychological safety first
- •Focus on systems, not blame
- •Balance critique with recognition
- •Drive to specific actions
- •Assign owners and dates to everything
Instructions
- •Ask about scope if not specified
- •Check prior context: Run
/relevant-learnings [topic]to see if we've learned similar lessons before - •Reference any project documents provided via @file syntax
- •Distinguish root causes from symptoms
- •Ensure every learning has an action
- •Save as markdown file
- •Offer to create presentation version using /present
Context Integration
After completing the retrospective:
- •Offer to save learnings: Ask "Should I save these learnings to the context registry? (
/context-save)" - •If yes, extract and save:
- •Each learning to
context/learnings/index.mdwith:- •Learning ID (L-NNN)
- •Clear one-sentence insight
- •Source (this retrospective)
- •Category (Strategy/Product/GTM/Customer/Process)
- •Confidence level
- •Application guidance
- •Each learning to
- •Update assumptions: If any assumptions were validated/invalidated, update
context/assumptions/registry.md - •Flag re-decisions: If findings trigger re-decision criteria for past decisions, highlight this