github / github/copilot-cli

/compact cannot recover a session after the CAPI Responses payload reaches the 5 MB limit`

Open
#4,424 0 comments 1 reaction 0 assignees View on GitHub
area:context-memory area:sessions
Dominant language
Shell
Stars
11.2k
Forks
1.9k
Avg merge
14h 16m
Merged PRs (30d)
6

Description

### Describe the bug

### Description

When a Copilot CLI session reaches the 5 MB CAPI Responses request limit, normal prompts fail as expected—but `/compact` also fails. This leaves no way to reduce the active context while preserving and continuing the session. Basically this kills my session and I cannot continue in it which is a huge kill in productivity and flow.

### Error

> The request is too large to send through CAPI Responses. Try shortening the conversation or prompt. (5.0 MB request; 5.0 MB limit)
>

### Environment

- GitHub Copilot CLI: `1.0.78-0`
- OS: macOS `26.5.2`
- Architecture: `arm64`

![Image](https://github.com/user-attachments/assets/8754091b-8195-4eb3-a940-ca3f2cb15f3f)

### Affected version

1.0.78-0

### Steps to reproduce the behavior

## Steps to reproduce

1. Continue an interactive Copilot CLI session until its request context reaches 5 MB.
2. Submit another prompt.
3. Observe the request-size error.
4. Run `/compact`.
5. Observe that compaction cannot complete either.

### Expected behavior

## Expected behavior

`/compact` should remain usable after the normal request limit is reached, perhaps by truncating or chunking context locally before requesting a summary. The complete transcript should remain stored locally.

Ideally, Copilot CLI should also compact automatically before reaching an unrecoverable size.

### Actual behavior

Both regular prompts and `/compact` are blocked. The session history still exists, but the session cannot be continued; the only practical workaround is starting a new session and manually transferring context.

### Additional context

Additional detail

This occurs when the UI reports only approximately 69–70% context usage (about 1.1M context shown), rather than near 100%.

The context meter appears to track token capacity, while CAPI applies a separate 5 MB serialized-request limit. The CLI neither warns about this lower effective limit nor compacts before reaching it.

I have attached a screenshot showing the session at 70% context when the failure occurred.

Contributor guide

Open the contributing guide

Research direction

Start by reproducing the failure in an interactive Copilot CLI session when the CAPI Responses request reaches the 5 MB limit, then run /compact. Trace the compaction and request-size handling paths; done means /compact can reduce context and continue the existing session, while the complete transcript remains stored locally.

Written by the indexing model from the issue text.

Assessment

Tech stack
shell
Domain
cli
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.