Compaction
The agent summarising its own context when it fills (auto-compact, /compact). When the context window fills, the agent replaces the conversation with its own summary and carries on from that.
Verdict
Keeps a run going, at the cost of the reasons; on long runs it compounds into drift. Use it, but never as your only memory.
Best for
Getting a single session past the window limit when nothing important has been said yet.
Poor fit
Runs measured in hours or days, where by the third compaction the agent works from a summary of a summary.
Scores
| Criterion | Score | Why |
|---|---|---|
| Survival across sessions | Weak | Exists only inside the running session. |
| Cross-tool reach | Weak | Each agent compacts in its own way, for itself. |
| Multi-agent sharing | Weak | Not shared. |
| Retention of reasons | Weak | Reasons are what a summary drops first. |
| Scale | Partial | Lets a run continue indefinitely, by losing detail every time. |
| Openness | Weak | The summarising prompt and what it keeps are decided by the vendor. |
Strengths
- Automatic and built in.
- Keeps the plan and the current state of the task.
Limits
- Drops the conversation: rejected options, constraints and who said what.
- Each compaction summarises the last summary, so losses compound (see /compaction-and-drift).
- Nothing leaves the session: other agents and later sessions never see it.
How it fails
- Hour 40: the agent re-introduces a library that was ruled out on hour 3, because the ruling was summarised away.
- A constraint stated once in chat ("sandboxes have no network") does not survive the first summary.
Cost
Included in the agent; the summarising call spends tokens from your plan or API bill.