# Smart Compact

> Compact the conversation context while preserving what matters most. Use this skill whenever the user runs /smart-compact or asks to "compact", "summarize and compact", or "clean up the context". French triggers: "compacte la conversation", "résume et compacte", "nettoie le contexte", "fais un résumé de la session". It generates a structured summary of the session — what was worked on, what changed, decisions made, and what's still pending — then triggers /compact with that summary as the guiding prompt so the compacted context retains the right information for the next session.

- Skill: `hugodorne/smart-compact` (Agent Skill)
- Install (CLI): `npx skillmds@latest add hugodorne/smart-compact`
- Raw SKILL.md: https://api.skillmd.com/api/skills/hugodorne/smart-compact/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: HugoDorne (https://skillmd.com/u/hugodorne)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/hugodorne/smart-compact

---


# Smart Compact

Analyze the current conversation, write a structured handoff summary, display it to
the user, and then trigger `/compact` with that summary as the compaction prompt.

## Summary structure

Build the summary using this template (omit sections that are empty):

```
## Session summary

### Context
<One or two sentences: the project, feature branch, or ticket being worked on.>

### Accomplished
<Bullet list of concrete things completed: files changed, bugs fixed, features added,
decisions made. Be specific — include file paths, class names, and the nature of the
change.>

### Key decisions
<Bullet list of non-obvious choices made and the reason behind each. Skip obvious things.>

### Current state
<One sentence: where things stand right now (e.g. "Tests pass, ready to commit" or
"Mid-implementation of X, stopping at Y").>

### Next steps
<Bullet list of what remains, in order of priority. Include blockers if any.>
```

## How to write a good summary

- Be specific: "Added `UniverseNextDepartureMapper` in `mapper/` to convert BO to API
  response" is better than "Updated a mapper".
- Capture the *why* of decisions: "Chose polling over WebSocket because the mobile
  client doesn't support long-lived connections" is future-gold.
- Keep it tight — the goal is to give the next session enough context to pick up
  without re-reading everything.

## Execution

1. Read through the full conversation to extract the relevant facts.
2. Write the summary using the template above.
3. Display it clearly to the user under a `## Conversation Summary` heading.
4. Immediately after displaying, invoke `/compact` passing the summary text as the
   compaction prompt so it guides what the compactor preserves.

The summary should be the *last thing* you show the user before `/compact` runs —
this way they can see what was captured before the context is trimmed.