Bug Description
Description:
When running the /compact command to reduce the context size, the operation reports a successful reduction in messages, but the token counter in the status bar incorrectly spikes to an unexpectedly high number.
In my session, the token count went from 56.9k before compaction to 1.1M immediately after, despite the logs stating that the token context was reduced by 0.0%.
Steps to Reproduce:
Start a session and build up a moderate amount of context (e.g., ~56.9k tokens).
Run the /compact command in the prompt.
Observe the console output and the updated status bar.
Expected Behavior:
The token count in the status bar should either decrease or remain relatively stable, accurately reflecting the remaining context after older messages are compacted.
Actual Behavior:
The CLI outputs: [09:54:08] Context size reduced by 0.0% (tokens), 92.4% (messages)
However, the status bar indicator jumps from FORGE 56.9k to FORGE 1.1M, showing a massive and incorrect increase in the token count.
Forge Version
2.11.3
Operating System & Version
macOs
AI Provider
z.ai
Model
glm-5.1
Installation Method
npm install -g forgecode
Configuration
Bug Description
Description:
When running the /compact command to reduce the context size, the operation reports a successful reduction in messages, but the token counter in the status bar incorrectly spikes to an unexpectedly high number.
In my session, the token count went from 56.9k before compaction to 1.1M immediately after, despite the logs stating that the token context was reduced by 0.0%.
Steps to Reproduce:
Start a session and build up a moderate amount of context (e.g., ~56.9k tokens).
Run the /compact command in the prompt.
Observe the console output and the updated status bar.
Expected Behavior:
The token count in the status bar should either decrease or remain relatively stable, accurately reflecting the remaining context after older messages are compacted.
Actual Behavior:
The CLI outputs: [09:54:08] Context size reduced by 0.0% (tokens), 92.4% (messages)
However, the status bar indicator jumps from FORGE 56.9k to FORGE 1.1M, showing a massive and incorrect increase in the token count.
Forge Version
2.11.3
Operating System & Version
macOs
AI Provider
z.ai
Model
glm-5.1
Installation Method
npm install -g forgecode
Configuration