aws / aws/amazon-q-developer-cli
Context summarization feature expands rather than compacts information
- Dominant language
- Rust
- Stars
- 2k
- Forks
- 439
- PR merge metrics
- No merged PRs in 30d
Description
### Checks
- [x] I have searched [github.com/aws/amazon-q-developer-cli/issues](https://github.com/aws/amazon-q-developer-cli/issues?q=) and there are no duplicates of my issue
- [x] I have run `q doctor` in the affected terminal session
- [x] I have run `q restart` and replicated the issue again
### Operating system
macOS 15.5.0 (24F74)
### Expected behaviour
The context summarization feature should compress conversation history into concise summaries that preserve essential information while reducing token usage. When Q says it's "compacting" information, it should actually make the content shorter and more efficient.
### Actual behaviour
The summarization feature creates verbose, expanded summaries with headers, bullet points, and repetitive information that are often longer than the original content. This defeats the purpose of summarization and contributes to context window bloat in longer conversations.
```
[chat-transcript]
You're absolutely right - that summary wasn't compacted at all. It actually expanded the information rather than condensing it. A proper compact summary should have been something like:
"Tested autosave functionality in save_session tool - confirmed working with proper tagging and daily counter tracking. ATOM-Quip workflow system validated as production-ready."
The original summary kept all the verbose details, bullet points, and repetitive informa
> what should we do about it? that is a native q functionality, isnt it?
You're right - this appears to be a native Q functionality for conversation context management. The verbose summary format suggests it's an automatic system feature rather than something we can directly control.
A few thoughts on what could be done:
1. **System-level improvement**: The summarization algorithm could be enhanced to actually compress information rather than just reformatting it with headers and bullet points.
2. **Configurable verbosity**: There could be settings to control summary detail level (brief/detailed/full).
3. **Smart filtering**: The system could identify truly relevant vs. redundant information and only preserve what's likely to be referenced again.
4. **Token efficiency**: Given context window limitations, more aggressive compression would be valuable for longer conversations.
Since this is built into Q's architecture, any improvements would need to come from the Q development team. The current implementation seems to prioritize completeness over conciseness, which has trade-offs - you get all the context but at the cost of efficiency.
Is this causing issues with context window limits in longer conversations, or is it more about the principle of ineffective "compacting"?
[Tool uses: none]
> both.
That's a real problem then. You're hitting both practical limitations (context window bloat) and the frustration of a feature that doesn't do what it claims to do.
For the immediate practical issue, you might need to:
- Start fresh conversations more frequently to avoid context overflow
- Be more selective about what gets saved to sessions if you have control over that
For the broader issue, this seems like something the Q team should address since:
1. "Compacting" that expands content is counterproductive
2. Context window management is critical for longer workflows
3. Users expect summarization to actually summarize
This feels like a case where the feature was designed with good intentions (preserve all context) but implemented without considering the real-world constraints (token limits) and user expectations (actual compression).
Have you noticed at what conversation length this becomes problematic? And is there any way to provide feedback to the Q development team about this specific issue?
[Tool uses: none]
> could you send this to the q developers?
I can help you report this issue to the Q developers using the built-in reporting tool. Let me create a GitHub issue for this:
[Tool uses: report_issue]
(...truncated)
```
### Steps to reproduce
1. Have a conversation with multiple tool uses and technical details
2. Allow the system to generate a context summary
3. Observe that the "compacted" summary is actually longer and more verbose than necessary
4. Notice impact on context window usage in longer conversations
### Environment
```yaml
[q-details]
version = "1.13.0"
hash = "7dee6efdf68f6343e80a56028e6b40f8eff77746"
date = "2025-07-31T02:57:26.581678Z (2d ago)"
[system-info]
os = "macOS 15.5.0 (24F74)"
chip = "Apple M1 Pro"
total-cores = 10
memory = "32.00 GB"
[environment]
cwd = "/Users/USER"
cli-path = "/Users/USER"
install-method = "unknown"
[env-vars]
PATH = "/Users/USER/.local/bin:/opt/homebrew/bin:/opt/homebrew/sbin:/Users/USER/.toolbox/bin:/usr/local/bin:/System/Cryptexes/App/usr/bin:/usr/bin:/bin:/usr/sbin:/sbin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/local/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/bin:/var/run/com.apple.security.cryptexd/codex.system/bootstrap/usr/appleinternal/bin:/usr/local/go/bin:/Users/USER/.local/bin:/Users/USER/.cargo/bin:/Applications/Visual Studio Code.app/Contents/Resources/app/bin:/opt/homebrew/opt/fzf/bin"
QTERM_SESSION_ID = "2579b8b0ed8d41d59c4133fc2f0a9ede"
Q_SET_PARENT_CHECK = "1"
Q_TERM = "1.13.0"
SHELL = "/bin/zsh"
TERM = "xterm-256color"
ZDOTDIR = "/Users/USER"
__CFBundleIdentifier = "com.microsoft.VSCode"
[chat-settings]
[chat-trusted_tools]
execute_bash=trusted
@q-session-manager/save_session=trusted
@amzn-mcp/create_quip=trusted
@amzn-mcp/edit_quip=trusted
fs_read=trusted
fs_write=trusted
[chat-failed_request_ids]
none
[chat-context]
current_profile=q_cli_default
profile_context=
AmazonQ.md
README.md
.amazonq/rules/**/*.md
files=none
```
Contributor guide
Assessment
This issue has not been assessed yet.