compress-session

Compress session conversation into structured summary for system memory. Activate when user says "compress", "compress session", or after significant work. Captures decisions, tools built, breakthroughs, and context that git commits can't preserve.

⭐5stars🔀1forks📁View on GitHub🕐Updated Jan 11, 2026

When & Why to Use This Skill

The compress-session skill is a specialized context optimization tool designed to transform lengthy AI conversations into structured, high-density summaries. Unlike standard logs, it captures the 'why' and 'how' behind technical decisions, breakthroughs, and tool development, ensuring that critical project context is preserved for long-term system memory and efficient session management.

Use Cases

  • Project Continuity: Maintaining a clear record of architectural decisions and logic shifts during multi-day development sprints to prevent context loss.
  • Token Window Management: Compressing exhaustive chat histories into concise summaries to stay within model context limits while retaining essential project milestones.
  • Technical Auditing: Documenting the rationale behind specific code changes and breakthroughs that git commits alone cannot preserve.
  • Session Handoffs: Generating a structured 'state of play' for users or agents returning to a complex task after a significant break or shift in focus.
namecompress-session
descriptionCompress session conversation into structured summary for system memory. Activate when user says "compress", "compress session", or after significant work. Captures decisions, tools built, breakthroughs, and context that git commits can't preserve.
allowed-toolsRead, Write, Bash

Compress Session

Compress active session into structured summary. Capture decisions, tools built, breakthroughs - the "why" and "how" that git commits don't preserve.

When to Activate

  • User says: "compress", "compress session"
  • After significant milestones
  • Every ~10 messages in long sessions
  • Before ending major work

Process

1. Determine Sequence Number

ls -t inbox/session-summaries/$(date +%Y-%m-%d)*.md 2>/dev/null | head -1
  • If files exist: increment (seq2, seq3...)
  • If none: this is seq1

2. Review Conversation

Identify signal vs. noise:

  • INCLUDE: Decisions, tools built, breakthroughs, key quotes, context
  • SKIP: Pleasantries, process discussion without outcome, redundant info

3. Write Summary

Create: inbox/session-summaries/YYYY-MM-DD-HHMM-seqN.md

# Session Capture: YYYY-MM-DD [Sequence N]

**Time:** [Start] - [Current]
**Roles Active:** [List or "general"]
**Messages:** [Approximate range]
**Branch:** [git branch]

---

## Context

[1-3 sentence narrative of what happened this sequence]

---

## Decisions Made

- [Decision 1]
  - Reason: [Why]
- [Decision 2]

---

## Tools/Workflows Built

- **[Name]:** [Description and purpose]

[Or: None this sequence]

---

## Breakthroughs/Insights

- [Insight] - [Significance]

---

## Key Quotes

- "[Quote]" ([Context])

---

## Files Created/Modified

[From git status]
- New: [files]
- Modified: [files]

---

## Next Steps

- [What's coming]

---

**Session compressed. Sequence [N] captured.**

4. Confirm

Session compressed. Sequence [N] captured.
Stored: inbox/session-summaries/YYYY-MM-DD-HHMM-seqN.md

What to Include

  • Decisions affecting work/strategy/direction
  • Tools/workflows built or planned
  • Breakthroughs (creative, technical, conceptual)
  • Context explaining why decisions were made
  • Key quotes revealing intent

What to Skip

  • Conversational filler
  • Detailed implementation (git shows that)
  • Process discussion without decision
  • Speculation that didn't become action

Tone

  • Concise, structured
  • Bullet points over paragraphs
  • Searchable and scannable
  • Neutral observation