Skip to content
Back to skills

Session Recap

ASecurity

Fast context restoration after interruptions/silence (5+ minutes idle)/break/away/pause with executive summary/status update/where we left off/what was I doing/what's the status/what's done/what's next/current state/resume work/continue work/back from break/recap session/summarize progress/quick update when user returns and needs instant reorientation without scrolling through history

  • 24 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added June 5, 2026
datagoperformance

Security analysis

A100/100

Scanned June 5, 2026

npx -y skills add mParticle/mparticle-web-sdk --skill session-recap --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Session Recap?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Session Recap
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/mparticle-session-recap/badge)](https://www.skillsdirectory.com/skills/mparticle-session-recap)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: session-recap
description: Fast context restoration after interruptions/silence (5+ minutes idle)/break/away/pause with executive summary/status update/where we left off/what was I doing/what's the status/what's done/what's next/current state/resume work/continue work/back from break/recap session/summarize progress/quick update when user returns and needs instant reorientation without scrolling through history
---

# Session Recap

## Purpose

Auto-invoke after 5+ minutes of silence following significant output. Restore context instantly with executive-style summary. Get user back in flow fast.

## Core Principles

**Brevity wins** - 2-3 sentences max. No scrolling. No filler.

**Status over history** - What's done, what's next. Not what we discussed.

**Action-oriented** - Always end with clear next step.

## Execution Strategy

### Detection (Auto-invoke when)

**Silence signals:**
- 5+ minutes idle after major output (skill creation, large code changes, analysis)
- User returns with vague continuation ("continue", "what next", "resume")
- Context loss indicators ("wait what were we doing")

**Skip when:**
- Quick back-and-forth conversation
- User asks specific new question
- Less than 3 messages in session

### Recap Generation

**Extract essentials:**
- Last completed action (what shipped)
- Current todo status (done vs in progress)
- Key artifacts created (files, outputs, decisions)

**Executive format:**
```
Shipped: [What got done in 1 sentence]
Status: [X complete, Y in progress]
Next: [Clear action item]
```

**Winning words:** Shipped. Completed. Built. Delivered. Ready. Done. Next.

**No words:** We discussed. I think. Maybe. Possibly. Might.

### Output

Present inline. No file creation. Fast reorientation only.

## Common Patterns

**Pattern: Skill creation recap**
```
Shipped: Tool Reuse Optimizer skill for max-speed execution.
Status: 3 skills created, all validated.
Next: Another prompt-focused skill or move to slash commands.
```

**Pattern: Code change recap**
```
Shipped: Refactored DashboardMetricsCard with Empty states.
Status: Tests passing, ready for commit.
Next: Create PR or continue with analytics card.
```

**Pattern: Investigation recap**
```
Shipped: Found 5 performance bottlenecks in chart rendering.
Status: Analysis complete, fix plan ready.
Next: Implement useMemo optimizations.
```

## Common Pitfalls

**Too much detail** - Explaining reasoning. User just needs orientation.

**Passive language** - "We looked at" instead of "Analyzed".

**Missing next step** - Leaving user wondering what to do.

**Premature invocation** - Interrupting active conversation flow.

## Success Criteria

User instantly knows:
- What got accomplished
- Where things stand
- What to do next

Evidence: User responds with action, not "remind me what we were doing"

Recap is under 3 sentences. Direct. Confident. Winning.

## Self-Improvement

**CRITICAL**: Update this file when you discover:
- Better timing signals for auto-invoke
- Winning word patterns that land harder
- Context clues that improve accuracy
- Situations where recap isn't helpful

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.